A method, device and filtering method for constructing an adaptive filtering model

Abstract

The present application provides a method, device and filtering method for constructing an adaptive filtering model. The method for constructing an adaptive filtering model provided in the present application can construct an adaptive filtering model, which can filter the real-time monitoring signal of the source device under the actual service state without the need for statistical characteristics of statistical signals and noise; in addition, the adaptive filtering model can not only enable the trained adaptive filtering model to effectively filter the interference signal at the working frequency of the noise source device by introducing modeling noise signals and random noise, but also enhance the denoising ability of the adaptive filtering model to random noise by introducing random noise; further, through adversarial training, the powerful learning ability and intelligent characteristics of the neural network can be used for continuous learning and experience accumulation, so that the trained adaptive filtering model can better adapt to the dynamic filtering needs under different environments.

Description

Technical Field

[0001] The present application relates to the field of filtering technology, and in particular to a method and device for constructing an adaptive filtering model and a filtering method. Background Art

[0002] With the rapid development of information technology, signal processing plays a vital role in many fields, such as communication, image and audio processing, biomedical engineering, etc. In these fields, it is often necessary to extract useful information from noisy signals, which makes filtering technology one of the key links.

[0003] Filtering technology has always been a key research direction in the field of signal processing. Typical filtering methods include median filtering, mean filtering, Kalman filtering, Wiener filtering, etc. Among them, median filtering is a nonlinear filtering method that removes noise by taking the median of each sample point in the signal within its neighborhood. Median filtering is effective in removing impulse noise such as salt and pepper noise, but has limited effect on other types of noise such as Gaussian noise. Mean filtering is also a nonlinear filtering method that removes noise by taking the mean of each sample point in the signal within its neighborhood. Mean filtering has a certain effect on removing Gaussian noise, but it will make the signal blurred. Kalman filtering is an optimal linear filtering method based on the state space model. It is suitable for state estimation and signal filtering of dynamic systems. It can estimate the state of the system and the parameters of the filter by recursion when the statistical characteristics of the noise are unknown. However, the computational complexity of Kalman filtering is high and its processing capacity for high-dimensional systems is limited. Wiener filtering is a linear filtering method based on the minimum mean square error criterion. It can achieve the best filtering effect when the statistical characteristics of the signal and noise are known. However, in practical applications, the statistical characteristics of signals and noise are often unknown, which limits the application of Wiener filtering.

[0004] As described above, the above-mentioned filtering methods based on signal analysis are highly dependent on the statistical characteristics of signals and noise, and require known statistical characteristics of signals and noise to achieve better filtering effects. However, in practical applications, these statistical characteristics are often unknown or time-varying, which greatly limits the performance of these algorithms. Summary of the invention

[0005] In view of this, the present application provides a method, device and filtering method for constructing an adaptive filtering model, so as to adaptively filter a noisy signal without requiring statistical characteristics of the signal and the noise.

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

[0007] The first aspect of the present application provides a method for constructing an adaptive filtering model, the method comprising:

[0008] Constructing an initial model based on adversarial neural network technology; wherein the initial model includes a generator and a discriminator;

[0009] Obtain multiple pure monitoring signals of source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on normal distribution;

[0010] For any noise source device on the physical device where the source device is located in the actual service state, model the noise caused by the vibration of the noise source device to obtain an interference noise model corresponding to the noise source device, and based on the interference noise model, obtain the modeled noise signal of the noise source device at each operating frequency to obtain the modeled noise signal corresponding to the noise source device;

[0011] Generate a noisy composite signal using any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device, and use the pure monitoring signal and the noisy composite signal as a training sample;

[0012] Based on the adversarial training method, the initial model is trained using multiple training samples to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device under actual service status. The second aspect of the present application provides a device for constructing an adaptive filtering model, the device comprising a construction module, an acquisition module and a processing module.

[0013] Wherein, the construction module is used to construct an initial model based on adversarial neural network technology; wherein, the initial model includes a generator and a discriminator;

[0014] The acquisition module is used to acquire multiple pure monitoring signals of the source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on a normal distribution;

[0015] The processing module is used to model the noise caused by the vibration of any noise source device on the physical device where the source device is located in the actual service state, obtain the interference noise model corresponding to the noise source device, and obtain the modeled noise signal of the noise source device at each operating frequency based on the interference noise model to obtain the modeled noise signal corresponding to the noise source device;

[0016] The processing module is further used to generate a noisy composite signal using any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device, and use the pure monitoring signal and the noisy composite signal as a training sample;

[0017] The processing module is also used to train the initial model using multiple training samples based on an adversarial training method to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device monitored in real time under actual service conditions. The method and device for constructing an adaptive filtering model provided in the present application construct an initial model including a generator and a discriminator based on the adversarial neural network technology, thereby obtaining multiple pure monitoring signals of a source device in a healthy state under various working conditions, and randomly generating multiple random noise signals based on a normal distribution, and then for any noise source device on the physical device where the source device is located in an actual service state, the noise caused by the vibration of the noise source device is modeled to obtain an interference noise model corresponding to the noise source device, and based on the interference noise model, the modeling noise signal of the noise source device at its various operating frequencies is obtained to obtain the modeling noise signal corresponding to the noise source device, and then any pure monitoring signal, any random noise signal, and any modeling noise signal corresponding to each noise source device is used to generate a noisy composite signal, and the pure monitoring signal and the noisy composite signal are used as a training sample. Furthermore, based on the adversarial training method, the initial model is trained using multiple training samples to obtain. A trained adaptive filtering model can, on the one hand, be trained to obtain an adaptive filtering model, and without the need for statistical characteristics of statistical signals and noise, the adaptive filtering model can be used to filter the real-time monitoring signal of the source device under actual service status; on the other hand, the adaptive filtering model introduces modeling noise signals and random noise, which can not only enable the trained adaptive filtering model to effectively filter the interference signal at the working frequency of the noise source device, but also enhance the denoising ability of the adaptive filtering model for random noise through the introduction of random noise, thereby improving the robustness of the adaptive filtering model; on the other hand, through adversarial training, the powerful learning ability and intelligent characteristics of the neural network can be used for continuous learning and experience accumulation. With the passage of network training time and the accumulation of data, the continuous evolution and upgrading of the filtering capability can be achieved, which can not only effectively improve the generator's processing capability for complex noise signals, but also enable the trained model to better adapt to the dynamic filtering needs under different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a first embodiment of a method for constructing an adaptive filtering model provided in the present application;

[0019] Figure 2 A schematic diagram of an initial model shown as an exemplary embodiment of the present application;

[0020] Figure 3 A schematic diagram of a generator shown as an exemplary embodiment of the present application;

[0021] Figure 4 A schematic diagram of a discriminator shown as an exemplary embodiment of the present application;

[0022] Figure 5 A flowchart of Embodiment 2 of the method for constructing an adaptive filtering model provided in this application;

[0023] Figure 6 A flowchart of a first embodiment of the filtering method provided in this application;

[0024] Figure 7 This is a structural diagram of Embodiment 1 of the device for constructing an adaptive filtering model provided in the present application. DETAILED DESCRIPTION

[0025] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0026] The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0027] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the 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".

[0028] Specific embodiments are given below to introduce the technical solution of the present application in detail.

[0029] Figure 1This is a flow chart of the first embodiment of the method for constructing an adaptive filtering model provided by this application. Figure 1 The method for constructing the adaptive filtering model provided in this embodiment may include:

[0030] S101. Construct an initial model based on adversarial neural network technology; wherein the initial model includes a generator and a discriminator.

[0031] Figure 2 This is a schematic diagram of an initial model shown in an exemplary embodiment of the present application. Figure 2 ,Adversarial neural network technology is a deep learning model architecture. The initial model built using this technology includes a generator and a discriminator.

[0032] Figure 3 This is a schematic diagram of a generator shown as an exemplary embodiment of the present application. Figure 4 This is a schematic diagram of a discriminator shown in an exemplary embodiment of the present application.

[0033] Please also refer to Figure 3 and Figure 4 , the generator includes an encoder and a decoder, and the discriminator includes an encoder. The generator is used to filter and reduce noise on the input data (the input data includes a noisy composite signal and a real-time monitoring signal, wherein in the training phase, the input data includes a noisy composite signal, and in the use phase, the input data is a real-time monitoring signal). Specifically, the encoder in the generator is used to extract features from the input data to obtain a feature map, and further, the decoder in the generator is used to reconstruct the signal based on the feature map to obtain a generated signal (the generated signal is the filtered signal after filtering).

[0034] Furthermore, during the training phase, the input data of the discriminator includes the pure monitoring signal and the generated signal generated by the generator. The discriminator judges the pure monitoring signal as true and the generated signal as false, and makes true and false judgments on the two types of signals.

[0035] S102. Acquire multiple pure monitoring signals of source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on a normal distribution.

[0036] It should be noted that the source device refers to the source device of the signal to be filtered, that is, the device from which the signal to be filtered comes is the source device. Specifically, the source device can be set according to actual needs and is not limited here. For example, in one possible implementation, the signal to be filtered is the vibration signal of the hydraulic pump on the aircraft. At this time, the signal to be filtered is a vibration signal (that is, the monitoring signal of the source device is a vibration signal), and the source device is a hydraulic pump. In addition, a pure monitoring signal refers to a signal that does not contain noise and interference.

[0037] Specifically, multiple pure monitoring signals of source equipment in a healthy state under various working conditions can be obtained through experimental methods or simulation methods.

[0038] It should be noted that in the process of obtaining pure monitoring signals, it is necessary to ensure the diversity and representativeness of the working conditions, so that the multiple pure monitoring signals obtained can cover all typical working conditions within the actual working envelope of the hydraulic pump as much as possible.

[0039] For example, in one possible implementation, when obtaining multiple pure monitoring signals of a source device in a healthy state under various working conditions through experimental methods, different working conditions can be set first according to the working envelope of the source device in actual service; further, for a selected source device in a healthy state, under each set working condition, the monitoring signal of the source device within a preset time length is collected; then, the collected monitoring signals are processed to remove noise and interference, so as to obtain multiple pure monitoring signals of the source device in a healthy state under various working conditions.

[0040] For another example, in another possible implementation, when obtaining multiple pure monitoring signals of a healthy source device under various working conditions through a simulation method, a simulation model of the source device can be established first; further, different working conditions can be set in the simulation model to simulate the behavior of the source device under various working conditions; then, multiple pure monitoring signals of the healthy source device under various working conditions can be output based on the simulation model.

[0041] It should be noted that, in specific implementation, a suitable method can be selected according to actual needs to obtain multiple pure monitoring signals of source equipment in a healthy state under various working conditions.

[0042] Furthermore, in this step, after obtaining multiple pure monitoring signals of the source equipment in a healthy state under various working conditions, multiple Gaussian white noise signals can be randomly generated based on a normal distribution and used as multiple random noise signals.

[0043] S103. For any noise source device on the physical device where the source device is located in the actual service state, model the noise caused by the vibration of the noise source device to obtain an interference noise model corresponding to the noise source device, and based on the interference noise model, obtain the modeled noise signal of the noise source device at its various operating frequencies to obtain the modeled noise signal corresponding to the noise source device.

[0044] It should be noted that the physical device is associated with the source device, and is the device where the source device is located in actual service, while the noise source device is a noise-causing device on the physical device other than the source device. Combined with the above example, when the source device is a hydraulic pump on an aircraft, at this time, the physical device is the aircraft, the noise source device is the engine, and the monitoring signal of the source device is a vibration signal.

[0045] Specifically, in this step, for a certain noise source device, it is necessary to first model the noise caused by the vibration of the noise source device to obtain an interference noise model corresponding to the noise source device.

[0046] In a specific implementation, the step of modeling the noise caused by the vibration of the noise source device to obtain the interference noise model corresponding to the noise source device may include:

[0047] (1) Under actual service conditions, the interference noise signal of the noise source equipment is collected in real time by sensors;

[0048] (2) analyzing the interference noise signal to obtain an analysis result; wherein the analysis result includes a time domain feature and a frequency domain feature;

[0049] (3) Based on the analysis results, an interference noise model of the noise source equipment is established.

[0050] It should be noted that after the interference noise signal is collected, it can be analyzed in the time domain and frequency domain to obtain time domain characteristics and frequency domain characteristics; among them, the time domain characteristics can reflect the change of the interference noise signal over time, and analyze the amplitude, mean, maximum value, standard deviation, etc. of the signal; further, the frequency shift characteristics can show the distribution of the interference noise signal at different frequencies. The interference noise signal can be converted into the frequency domain through Fourier transform or other signal processing methods, and its spectrum can be observed to analyze the main frequency, harmonic components and other characteristics of the signal.

[0051] Furthermore, based on the analysis results, combined with the working state of the noise source equipment and the characteristics of the noise source, an interference noise model of the noise source equipment is established, and the interference noise model can be used to predict the noise behavior of the noise source equipment under different working conditions.

[0052] Referring to the foregoing description, it can be understood that the analysis results including the time domain characteristics and the frequency domain characteristics can provide a comprehensive understanding of the interference noise signal, so that the interference noise model of the noise source device finally established is more accurate.

[0053] Furthermore, after the interference noise model is obtained, the modeled noise signal of the noise source device at each operating frequency can be obtained based on the interference noise model.

[0054] It should be noted that the operating frequency refers to the frequency of the noise source device when it is actually in service. In combination with the above example, for example, when the noise source device is an engine, its operating frequency is the frequency of the engine when it is working on the aircraft.

[0055] S104, using any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device to generate a noisy composite signal, and using the pure signal and the noisy composite signal as a training sample.

[0056] Specifically, any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device may be added or multiplied to obtain the noisy composite signal.

[0057] In specific implementation, a first database can be established, and all pure monitoring signals can be stored in the first database; further, for each noise source device, a corresponding second database can be established for it, and all modeled noise signals corresponding to the noise source device can be stored in the second database. In addition, a third database can be established, and all random noise signals can be stored in the third database. In this step, a pure monitoring signal can be randomly selected from the first database, and then a modeled noise signal can be selected from the second data path corresponding to each noise source device, and a random noise signal can be selected from the third database, and then the selected pure monitoring signal, the modeled noise signal corresponding to each noise source device, and the random noise signal are additively superimposed or multiplicatively superimposed to obtain a noisy composite signal, and the pure monitoring signal and the noisy composite signal are used as a training sample.

[0058] It should be noted that additive superposition refers to the numerical addition of the selected pure monitoring signal, modeling noise signal and random noise signal; while multiplicative superposition refers to the numerical multiplication of the selected pure monitoring signal, modeling noise signal and random noise signal.

[0059] S105. Based on an adversarial training method, the initial model is trained using a plurality of training samples to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device under actual service status.

[0060] In a specific implementation, in a possible implementation, the training process of the initial model may include:

[0061] (1) Initialize the parameters of the generator and discriminator;

[0062] (2) training the generator; wherein, when training the generator, inputting the noisy composite signal into the generator, so that the generator filters and reduces noise on the noisy composite signal, and outputs a generated signal corresponding to the noisy composite signal; the generator optimizes parameters using a distance error between the generated signal corresponding to the noisy composite signal and a pure monitoring signal corresponding to the noisy composite signal as a loss;

[0063] (3) training the discriminator; wherein, when training the discriminator, the generated signal corresponding to the noisy composite signal is used as a negative sample, and the pure monitoring signal corresponding to the noisy composite signal is used as a positive sample, and parameter optimization is performed by minimizing the loss function of the discriminator;

[0064] (4) Repeat the steps of training the generator and training the discriminator until the preset number of training times is reached or the loss function converges.

[0065] It should be noted that in the initial stage of training, the generator and discriminator need to set initial parameters. In specific implementation, random initialization methods are usually used to set weights and bias values ​​to provide a basis for subsequent model training.

[0066] Further, see Figure 2 When training the generator, a noisy composite signal is input, and the generator attempts to filter and denoise the noisy signal and output a generated signal that is as close to the pure signal as possible. In addition, the generator optimizes its parameters by calculating the error (such as mean square error) between the generated signal and the pure monitoring signal to reduce this error. When training the discriminator, the discriminator regards the generated signal output by the generator as a negative sample and the original pure monitoring signal as a positive sample. Furthermore, the goal of the discriminator is to distinguish between negative and positive samples as accurately as possible, and to optimize its parameters by minimizing the loss function of the discriminator.

[0067] In addition, the training of the generator and the discriminator is carried out alternately, and the parameters are updated after each round of training. The training ends when the preset number of training times is reached or when the change of the loss function tends to be stable.

[0068] It should be noted that the loss functions of the generator and the discriminator are set according to actual needs and are not limited here. For example, in one embodiment, the generator can use mean square error as the loss function, and the discriminator can use cross entropy as the loss function.

[0069] It is understandable that through adversarial training, the generator and the discriminator are constantly competing with each other. The goal of the generator is to generate a signal as close to the real and pure as possible from the noisy signal, while the task of the discriminator is to distinguish the generated signal from the real signal. During the training process, the generator extracts features from the noisy signal and tries to denoise it to generate a generated signal similar to the pure signal; the discriminator learns how to distinguish the generated signal from the original pure signal. As the training progresses, the generator will gradually learn to better extract useful features from complex noise environments and achieve a more efficient noise reduction effect. At the same time, the feedback from the discriminator prompts the generator to continuously improve, thereby improving the quality of the generated signal until the generated signal is difficult to distinguish from the pure signal. In this way, the powerful learning ability and intelligent characteristics of the neural network can be used for continuous learning and experience accumulation. With the passage of network training time and the accumulation of data, the filtering ability is continuously evolved and upgraded, effectively improving the generator's ability to process complex noise signals, and it has strong adaptability and can better adapt to the dynamic filtering needs of different environments and different hydraulic pumps.

[0070] In addition, the method for constructing an adaptive filtering model provided by the present application, by introducing a modeling noise signal, after model training, the trained adaptive filtering model can effectively filter out the interference signal at the working frequency of the noise source device; at the same time, by introducing a random noise signal, the random noise signal can be introduced during the training process of the adaptive filtering model, which can enhance the denoising ability of the adaptive filtering model for random noise. In this way, when the working state of the source device changes and the interference signal of the noise source device changes dynamically, the monitored interfered signal can be adaptively filtered without manually adjusting the filtering parameters, and the robustness is good.

[0071] Further, referring to the above description, it can be understood that the trained adaptive filtering model can be subsequently used to filter the real-time monitoring signal of the source device under actual service status.

[0072] The method for constructing an adaptive filter provided in the present embodiment constructs an initial model including a generator and a discriminator based on the adversarial neural network technology, thereby obtaining multiple pure monitoring signals of the source device in a healthy state under various working conditions, and randomly generates multiple random noise signals based on a normal distribution, and then for any noise source device on the physical device where the source device is located in an actual service state, the noise caused by the vibration of the noise source device is modeled to obtain an interference noise model corresponding to the noise source device, and based on the interference noise model, the modeling noise signal of the noise source device at its various operating frequencies is obtained to obtain the modeling noise signal corresponding to the noise source device, and then any pure monitoring signal, any random noise signal, and any modeling noise signal corresponding to each noise source device is used to generate a noisy composite signal, and the pure monitoring signal and the noisy composite signal are used as a training sample. Further, based on the adversarial training method, the initial model is trained using multiple training samples to obtain a training A good adaptive filtering model, in this way, on the one hand, an adaptive filtering model can be trained to obtain a real-time monitoring signal of the source equipment under actual service status without the need for statistical characteristics of statistical signals and noise. The adaptive filtering model can be used to filter the real-time monitoring signal of the source equipment under actual service status in real time. On the second hand, the adaptive filtering model introduces modeling noise signals and random noise, which can not only enable the trained adaptive filtering model to effectively filter out the interference signal at the working frequency of the noise source equipment, but also enhance the denoising ability of the adaptive filtering model to random noise through the introduction of random noise, thereby improving the robustness of the adaptive filtering model. On the third hand, through adversarial training, the powerful learning ability and intelligent characteristics of the neural network can be used for continuous learning and experience accumulation. With the passage of network training time and the accumulation of data, the continuous evolution and upgrading of the filtering capability can be achieved, which can not only effectively improve the generator's processing capability for complex noise signals, but also enable the trained model to better adapt to the dynamic filtering needs under different environments.

[0073] Figure 5 This is a flowchart of the second embodiment of the method for constructing an adaptive filtering model provided by this application. Please also refer to Figure 5 , the method for constructing the adaptive filtering model provided in this embodiment, after obtaining the trained adaptive filtering model, the method further includes:

[0074] S501. Acquire a real-time monitoring signal of the source device under actual service status.

[0075] S502: Input the real-time monitoring signal into the adaptive filtering model, so that the generator in the adaptive filtering model filters and reduces noise on the real-time monitoring signal, and outputs a filtered signal.

[0076] Specifically, referring to the previous description, the encoder in the generator can perform feature extraction on the real-time detection signal to obtain a feature map. Furthermore, the decoder reconstructs the signal based on the feature map to obtain a filtered signal after filtering.

[0077] S503: Evaluate the filtered signal to obtain an evaluation result.

[0078] Specifically, in order to determine the filtering effect of the trained adaptive filtering model, the filtered signal can be evaluated. The evaluation criteria for the filtered signal can be selected according to actual needs and are not limited here. For example, in one embodiment, the signal-to-noise ratio can be selected as the evaluation result.

[0079] S504: judging whether the filtering effect of the adaptive filtering model meets the requirements based on the evaluation result, and optimizing the parameters of the adaptive filtering model or retraining the adaptive filtering model when the filtering effect of the adaptive filtering model does not meet the requirements.

[0080] Specifically, in this step, the evaluation result of the filtered signal can be judged according to the selected evaluation criteria to determine whether the filtering effect of the adaptive filtering model meets the requirements. For example, if the signal-to-noise ratio is selected as the evaluation criterion, a signal-to-noise ratio threshold can be preset. If the evaluation result is lower than the preset signal-to-noise ratio threshold, it is considered that the filtering effect does not meet the requirements. If the evaluation result is higher than the preset signal-to-noise ratio threshold, it is considered that the filtering effect meets the requirements.

[0081] Furthermore, if it is determined from the obtained evaluation results that the filtering effect of the adaptive model does not meet the requirements, in this step, the parameters of the adaptive filtering model can be further optimized or the adaptive filtering model can be retrained to improve the adaptive filtering model.

[0082] The method for constructing an adaptive filtering model provided in the present embodiment obtains a real-time monitoring signal of a source device under actual service status, and inputs the real-time monitoring signal into the adaptive filtering model, so that a generator in the adaptive filtering model filters and reduces noise on the real-time monitoring signal, outputs a filtered signal, and then evaluates the filtered signal according to a preset evaluation standard to obtain an evaluation result, thereby judging whether the filtering effect of the adaptive filtering model meets the requirements based on the obtained evaluation result, and when the filtering effect of the adaptive filtering model does not meet the requirements, optimizing the parameters of the adaptive filtering model or retraining the adaptive filtering model, so that the performance of the filter can be kept stable to avoid performance degradation in long-term operation.

[0083] The second aspect of the present application also provides a filtering method, and the filtering method provided by the present application is introduced below:

[0084] Figure 6 Please refer to the flowchart of the first embodiment of the filtering method provided in this application. Figure 6 The method is implemented based on any adaptive filtering model constructed according to the first aspect of the present application, and the method comprises:

[0085] S601. Acquire a real-time monitoring signal of a source device under actual service status.

[0086] S602: Input the real-time monitoring signal into the adaptive filtering model, so that the generator in the adaptive filtering model filters and reduces noise on the real-time monitoring signal, and outputs a filtered signal.

[0087] The filtering method provided in the present application adopts an adversarial neural network training architecture to conduct adversarial training on the generator and the discriminator, and realizes feature extraction and noise reduction of noisy signals through continuous game between the generator and the discriminator; in addition, by mixing the modeling noise signal, the random noise signal and the pure monitoring signal, the controllable regulation and generalization enhancement of the filtering characteristics of the adaptive filtering model can be achieved at the same time, thereby enhancing the filtering effect of the adaptive filtering model.

[0088] Corresponding to the aforementioned embodiment of a method for constructing an adaptive filtering model, the present application also provides an embodiment of a device for constructing an adaptive filtering model. Figure 7 This is a schematic diagram of the structure of the first embodiment of the construction device of the adaptive filtering model provided by the present application. Figure 7 The device provided in this embodiment includes a construction module 710, an acquisition module 720 and a processing module 730.

[0089] The construction module 710 is used to construct an initial model based on the adversarial neural network technology; wherein the initial model includes a generator and a discriminator;

[0090] The acquisition module 720 is used to acquire multiple pure monitoring signals of the source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on a normal distribution;

[0091] The processing module 730 is used to model the noise caused by the vibration of any noise source device on the physical device where the source device is located in the actual service state, obtain the interference noise model corresponding to the noise source device, and obtain the modeled noise signal of the noise source device at each operating frequency based on the interference noise model to obtain the modeled noise signal corresponding to the noise source device;

[0092] The processing module 730 is further used to generate a noisy composite signal using any clean monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device, and use the clean monitoring signal and the noisy composite signal as a training sample;

[0093] The processing module 730 is also used to train the initial model using multiple training samples based on an adversarial training method to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device under actual service status.

[0094] 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.

[0095] 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.

[0096] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only 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.

[0097] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, 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 constructing an adaptive filtering model, characterized in that: The method comprises: Constructing an initial model based on adversarial neural network technology; wherein the initial model includes a generator and a discriminator; Obtain multiple pure monitoring signals of source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on normal distribution; For any noise source device on the physical device where the source device is located in the actual service state, model the noise caused by the vibration of the noise source device to obtain an interference noise model corresponding to the noise source device, and based on the interference noise model, obtain the modeled noise signal of the noise source device at each operating frequency to obtain the modeled noise signal corresponding to the noise source device; Generate a noisy composite signal using any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device, and use the pure monitoring signal and the noisy composite signal as a training sample; Based on the adversarial training method, the initial model is trained using multiple training samples to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device monitored in real time under actual service conditions; wherein the source device is a hydraulic pump, the physical device is an aircraft, the noise source device is an engine, and the monitoring signal of the source device is a vibration signal.

2. The method according to claim 1, characterized in that The step of modeling the noise caused by the vibration of any noise source device on the physical device where the source device is located in the actual service state to obtain the interference noise model corresponding to the noise source device includes: Under actual service conditions, the interference noise signal of the noise source equipment is collected in real time by sensors; Analyze the interference noise signal to obtain analysis results; wherein the analysis results include time domain characteristics and frequency domain characteristics; Based on the analysis result, an interference noise model of the noise source device is established.

3. The method according to claim 1, characterized in that The method of generating a noisy composite signal by using any one of the pure monitoring signals, any one of the random noise signals, and any one of the modeled noise signals corresponding to each noise source device comprises: Any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device are added or multiplied to obtain the noisy composite signal.

4. The method according to claim 1, characterized in that: The adversarial training method is based on using a plurality of training samples to train the initial model to obtain an adaptive filtering model, including: Initialize the parameters of the generator and the discriminator; The generator is trained; wherein, when the generator is trained, the noisy composite signal is input into the generator, so that the generator filters and reduces noise on the noisy composite signal, and outputs a generated signal corresponding to the noisy composite signal; the generator optimizes parameters using the distance error between the generated signal corresponding to the noisy composite signal and the pure monitoring signal corresponding to the noisy composite signal as the loss; The discriminator is trained; wherein, when the discriminator is trained, a generated signal corresponding to the noisy composite signal is used as a negative sample, and a pure monitoring signal corresponding to the noisy composite signal is used as a positive sample, and parameter optimization is performed by minimizing a loss function of the discriminator; The steps of training the generator and training the discriminator are repeated until the preset number of training times is reached or the loss function converges.

5. The method according to claim 1, characterized in that: After obtaining the trained adaptive filtering model, the method further includes: Acquire a real-time monitoring signal of the source device under actual service status; Inputting the real-time monitoring signal into the adaptive filtering model, so that the generator in the adaptive filtering model filters and reduces noise on the real-time monitoring signal, and outputs a filtered signal; Evaluating the filtered signal to obtain an evaluation result; Based on the evaluation result, it is judged whether the filtering effect of the adaptive filtering model meets the requirements, and when the filtering effect of the adaptive filtering model does not meet the requirements, the parameters of the adaptive filtering model are optimized or the adaptive filtering model is retrained.

6. The method according to claim 1, characterized in that The method of obtaining multiple pure monitoring signals of the source equipment in a healthy state under various working conditions includes: Through experimental methods or simulation methods, multiple pure monitoring signals of source equipment in a healthy state under various working conditions are obtained.

7. The method according to claim 6, characterized in that The experimental method is used to obtain multiple pure monitoring signals of the source equipment in a healthy state under various working conditions, including: Setting different working conditions according to the working envelope of the source equipment in actual service state; For a selected source device in a healthy state, under each set working condition, a monitoring signal of the source device is collected within a preset time period; Process the collected monitoring signals to remove noise and interference, and obtain multiple pure monitoring signals of the source equipment in a healthy state under various working conditions; or, The method of obtaining multiple pure monitoring signals of the healthy source equipment under various working conditions by simulation includes: establishing a simulation model of the source device; Setting different operating conditions in the simulation model to simulate the behavior of the source device under various operating conditions; Based on the simulation model, multiple pure monitoring signals of the source equipment in health status under various working conditions are output.

8. A filtering method, characterized in that: The method is implemented based on an adaptive filtering model, and the adaptive filtering model is constructed based on the method for constructing an adaptive filtering model according to any one of claims 1 to 7; the method comprises: Obtain real-time monitoring signals of source equipment under actual service conditions; The real-time monitoring signal is input into the adaptive filtering model, so that the generator in the adaptive filtering model filters and reduces noise on the real-time monitoring signal, and outputs a filtered signal.

9. A device for constructing an adaptive filtering model, characterized in that: The device comprises a construction module, an acquisition module and a processing module, wherein: The building module is used to build an initial model based on the adversarial neural network technology; wherein the initial model includes a generator and a discriminator; The acquisition module is used to acquire multiple pure monitoring signals of the source equipment in a healthy state under various working conditions, and randomly generate multiple random noise signals based on a normal distribution; The processing module is used to model the noise caused by the vibration of any noise source device on the physical device where the source device is located in the actual service state, obtain the interference noise model corresponding to the noise source device, and obtain the modeled noise signal of the noise source device at each operating frequency based on the interference noise model to obtain the modeled noise signal corresponding to the noise source device; The processing module is further used to generate a noisy composite signal using any pure monitoring signal, any random noise signal, and any modeled noise signal corresponding to each noise source device, and use the pure monitoring signal and the noisy composite signal as a training sample; The processing module is also used to train the initial model using multiple training samples based on an adversarial training method to obtain a trained adaptive filtering model; wherein the adaptive filtering model is used to filter the real-time monitoring signal of the source device monitored in real time under actual service conditions.

Images