Vibration signal denoising method and device
By slicing and frequency domain feature extraction of vibration signals and dynamically adjusting the denoising parameters, the problem of difficulty in removing complex and variable vibration signal noise in the prior art is solved, and efficient and accurate denoising effect is achieved.
Patent Information
- Application Number
- CN202510250770.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-04
- Publication Date
- 2025-07-01
AI Technical Summary
The prior art is difficult to effectively remove noise in complex and variable vibration signals, resulting in a degradation in data quality and impact on signal recognition accuracy.
By slicing the vibration signal and extracting the frequency domain characteristics of the sliced signal, dynamically adjusting the denoising parameters, and optimizing the denoising model to improve the denoising effect.
It realizes rapid and accurate noise denoising of complex vibration signals, improving data quality and signal recognition accuracy.
Smart Images
Figure CN120234540A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vibration signal processing, and particularly to a vibration signal denoising method and device. Background Art
[0002] In the field of vibration signal monitoring, the quality of vibration signals is crucial for the accurate identification of underground targets. However, in practical applications, vibration signals are often interfered by various noises, including random noise, instrument noise, environmental noise, etc. These noises not only reduce the signal-to-noise ratio of the data, but also affect the accuracy of signal recognition and interpretation. Especially random noise, its complexity and randomness pose great challenges to data processing. Therefore, how to effectively remove the noise in vibration signals and improve the data quality has become an important research direction in the field of vibration signal processing.
[0003] Traditional denoising methods mainly rely on specific attributes of vibration signals, and a series of algorithms are used to extract effective signals and suppress noises, such as linear or quasi-linear prediction filters. These methods usually assume that vibration signals have a certain regularity. However, in actual situations, vibration signals often contain complex non-linear characteristics, making these methods difficult to effectively handle. Especially when facing complex and variable vibration signals, there are obvious limitations. Denoising methods based on the spatial domain, transform domain, and hybrid domain use the correlation between pixels in images or data to remove noises. Hybrid domain denoising methods usually need to execute different denoising algorithms in multiple domains, and hybrid domain denoising methods usually contain multiple parameters. The selection of these parameters is crucial for the denoising effect. However, the selection of traditional denoising parameters depends on manual experience for debugging and threshold setting. If the parameter selection is inappropriate, it may lead to poor processing effects, greatly reducing the efficiency and accuracy of denoising.
[0004] Therefore, there is an urgent need for a vibration signal denoising method that can quickly and accurately denoise complex and variable vibration signals. Summary of the Invention
[0005] In view of this, it is necessary to provide a vibration signal denoising method and device, which slice the vibration signal, extract the frequency domain features of the sliced signals, and dynamically adjust the denoising parameters according to the frequency domain features to improve the accuracy and efficiency of vibration signal denoising.
[0006] To solve the above technical problems, on the one hand, the present invention provides a vibration signal denoising method, including: Determine the initial denoised sliced signal of the sliced signals, the frequency domain features of the sliced signals, and the frequency domain features of the initial denoised sliced signal according to multiple consecutive sliced signals of the target vibration signal; Determine the noise spectrum of each slice signal according to the frequency-domain characteristics of the slice signal and the frequency-domain characteristics of the initial denoised slice signal, and iteratively optimize the denoising parameters according to the noise spectrum to obtain the optimal denoising parameters; Denoise and reconstruct the slice signal according to the optimal denoising parameters to obtain the denoised signal of the target vibration signal.
[0007] In a possible implementation manner, determining the initial denoised slice signal, the frequency-domain characteristics of the slice signal, and the frequency-domain characteristics of the initial denoised slice signal according to multiple consecutive slice signals of the target vibration signal includes: Construct a denoising model based on DnCNN, where the input of the denoising model is the noisy vibration signal and the output is the denoised signal; Slice the target vibration signal based on the sliding window method to obtain multiple consecutive slice signals; Obtain the initial denoised slice signal according to the denoising model and the slice signal; Extract the frequency-domain characteristics of the slice signal and the initial denoised slice signal based on the multi-scale feature extraction algorithm.
[0008] In a possible implementation manner, constructing a denoising model includes: Construct an initial model with the noisy vibration signal as the input and the denoised signal as the output, where the initial model includes a denoising task and a feature extraction task; Constrain the output of the initial model according to the physical rules of the vibration signal to obtain a constrained model; Construct a loss function according to the denoising task and the feature extraction task; Iteratively optimize and adjust the constrained model based on the reinforcement learning method and the loss function to obtain the denoising model.
[0009] In a possible implementation manner, the loss function of the denoising model is: , where is the denoising task loss value, is the feature extraction task loss value, is the model loss function value, is a constant, and the weights of the vibration signal denoising task and the feature extraction task are balanced by this constant.
[0010] In a possible implementation manner, determining the noise spectrum of each slice signal according to the frequency-domain characteristics of the slice signal and the frequency-domain characteristics of the initial denoised slice signal, and iteratively optimizing the denoising parameters according to the noise spectrum to obtain the optimal denoising parameters includes: Construct an optimization function for the denoising parameters of the denoising model according to the noise power spectral density; Determine the noise spectrum of the slice signal according to the frequency-domain characteristics of the slice signal and the initial denoised slice signal; Determine the noise power spectral density of the slice signal according to the noise spectrum; Based on the denoising parameter optimization function, iteratively optimize the denoising parameters according to the noise power spectral density to obtain the optimal denoising parameters.
[0011] In a possible implementation manner, the calculation formula of the noise spectrum is: , where is the frequency-domain characteristic of the slice signal at the moment, is the frequency-domain characteristic of the denoised slice signal, is the noise spectrum of the slice signal, is the window time length of the slice signal.
[0012] In a possible implementation manner, based on the denoising parameter optimization function, iteratively optimize the denoising parameters according to the noise power spectral density to obtain the optimal denoising parameters, including: Based on an optimization algorithm, adjust the denoising parameters of the denoising model according to the noise power spectral density; After multiple iterations, when the index value output by the denoising parameter optimization function reaches a preset convergence condition, stop the iteration to obtain the optimal denoising parameters.
[0013] In a possible implementation manner, denoise and reconstruct the slice signal according to the optimal denoising parameters to obtain the denoised signal of the target vibration signal, including: Adjust the hyperparameters of the denoising model according to the optimal denoising parameters to obtain the optimal denoising model; Use the original slice signal as the input of the optimal denoising model to output the optimal denoised slice signal; Reconstruct multiple optimal denoised slice signals according to the time-domain information to obtain the denoised signal of the target vibration signal.
[0014] In a possible implementation manner, the denoising model includes an input layer, a denoising layer, a multi-scale feature extraction layer, multiple hidden layers, and an output layer; The input layer is used to receive the input target vibration signal; The multi-scale feature extraction layer is used to capture the multi-scale information of the vibration signal through multiple convolutional layers and extract the frequency characteristics and time characteristics; The hidden layer is used to extract deep features from different levels; The output layer is used to output the denoised vibration signal.
[0015] Second aspect, the present invention further provides a vibration signal denoising device, including: A feature extraction module, configured to determine an initial denoised slice signal of the slice signals, a frequency domain feature of the slice signals, and a frequency domain feature of the initial denoised slice signal according to a plurality of consecutive slice signals of a target vibration signal; A denoising parameter optimization module, configured to determine a noise spectrum of each slice signal according to the frequency domain feature of the slice signals and the frequency domain feature of the initial denoised slice signal, and iteratively optimize the denoising parameters according to the noise spectrum to obtain optimal denoising parameters; A denoising module, configured to perform denoising reconstruction on the slice signals according to the optimal denoising parameters to obtain a denoised signal of the target vibration signal.
[0016] The beneficial effects of the present invention are as follows: First, slicing the target vibration signal can decompose the complex vibration signal into slice signals in a smaller range, and can more accurately locate and analyze the characteristics of the vibration signal; then, performing preliminary denoising processing on the sliced vibration signal, and extracting the frequency domain features of the slice signals before and after denoising, and displaying the energy distribution and intensity of the slice signals in different frequency bands through the extraction of the frequency domain features; then, determining the noise spectrum of the slice signals according to the frequency domain features of the slice signals before and after denoising, and accurately obtaining the noise intensity in different frequency bands in the frequency domain, analyzing the distribution and intensity of the noise at different frequencies according to the noise spectrum, and dynamically optimizing and adjusting the initial denoised signal; finally, using the optimized denoising parameters to perform denoising reconstruction on the slice signals again to obtain the denoised signal of the final target vibration signal. The present invention extracts features from the sliced vibration signals, dynamically adjusts the denoising parameters according to the different frequency domain features of each slice signal, adaptively optimizes the denoising performance, enhances the denoising ability for complex signals, and improves the accuracy and efficiency of vibration signal denoising. Description of the Drawings
[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can be obtained according to these drawings.
[0018] Figure 1 It is a schematic flowchart of an embodiment of the vibration signal denoising method provided by the present invention; Figure 2 For the present invention Figure 1 It is a schematic flowchart of an embodiment of step S101 in the present invention; Figure 3 For the present invention Figure 1 It is a schematic flowchart of an embodiment of step S102 in the present invention; Figure 4 For an embodiment of step S103 in the present invention Figure 1 The schematic flow chart is shown; Figure 5 For an embodiment of step S103 in the present invention Figure 2 The schematic flow chart is shown; Figure 6 For an embodiment of step S103 in the present invention Figure 3 The schematic flow chart of an embodiment before step S304 is shown; Figure 7 The schematic structural diagram of an embodiment of the vibration signal denoising device provided by the present invention is shown. Detailed implementation manners
[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.
[0020] In the description of the embodiments of the present invention, unless otherwise specified, "a plurality of" means two or more.
[0021] The descriptions such as "first" and "second" involved in the embodiments of the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one of such features.
[0022] Referring to "embodiments" herein means that the specific features, structures, or characteristics described in connection with the embodiments may be included in at least one embodiment of the present invention. The phrase appears in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.
[0023] The present invention provides a vibration signal denoising method and device, which will be described separately below.
[0024] Figure 1 For an embodiment flow chart of the vibration signal denoising method provided by the present invention, as Figure 1 shown, the vibration signal denoising method includes: S101. Determine the initial denoised slice signal of the slice signal, the frequency domain characteristics of the slice signal, and the frequency domain characteristics of the initial denoised slice signal according to a plurality of consecutive slice signals of the target vibration signal; It should be noted that vibration signals are obtained through sensors such as vibration sensors and sent to a signal processing terminal. Here, the signal processing terminal is an electronic terminal with data processing capabilities, such as a PC, a mobile terminal, a portable terminal, etc. Specifically, the signal processing software of the signal processing terminal (such as an APP on the mobile phone side and a client on the PC side) is used to perform sliding window slicing, denoising, and feature extraction on the vibration signals.
[0025] Among them, the method of slicing the vibration signals includes, but is not limited to, slicing the vibration signals by using the sliding window method. The window size and the sliding step length need to be kept consistent to ensure the matching of the slices of the samples and labels. The choice of the denoising method is not limited to the DnCNN network. Other network models, wavelet transforms, filters, or other deep learning algorithms can also be used to denoise the sliced vibration signals. The methods of feature extraction include, but are not limited to, multi-scale feature extraction algorithms. The fast Fourier transform (FFT) can also be used to extract the frequency domain features of the sliced vibration signals before and after denoising.
[0026] S102. Determine the noise spectrum of each sliced signal according to the frequency domain features of the sliced signal and the frequency domain features of the initially denoised sliced signal, and iteratively optimize the denoising parameters according to the noise spectrum to obtain the optimal denoising parameters. It should be noted that by comparing the frequency domain features of the sliced signal before and after denoising, it is possible to identify which frequency components are noise. Specifically, if a certain frequency component is significantly weakened or disappears after initial denoising, then it is very likely to be a noise component. According to these identified noise components, determine the noise spectrum of each sliced signal. The noise spectrum describes the distribution and intensity of the noise at different frequencies, and use the noise spectrum distribution and intensity to select appropriate denoising parameters.
[0027] S103. Denoise and reconstruct the sliced signal according to the optimal denoising parameters to obtain the denoised signal of the target vibration signal.
[0028] It should be noted that the target vibration signal contains time domain information. During the process of slicing, denoising, and feature extraction of the target vibration signal, its time domain information is retained. Finally, the multiple sliced vibration signals denoised with the optimal denoising parameters are recombined in the time order of the original signal to obtain the final denoised vibration signal.
[0029] Compared with the prior art, in this embodiment, the target vibration signal is sliced, and the complex vibration signal can be decomposed into sliced signals in a smaller range, enabling more accurate localization and analysis of the characteristics of the vibration signal. Then, the sliced vibration signal is preliminarily denoised, and the frequency-domain characteristics of the sliced signal before and after denoising are extracted. The energy distribution and intensity of the sliced signal in different frequency bands are shown through the extraction of the frequency-domain characteristics. Then, according to the frequency-domain characteristics of the sliced signal before and after denoising, the noise spectrum of the sliced signal is determined, and the noise intensity in different frequency bands in the frequency domain can be accurately obtained. The distribution and intensity of the noise at different frequencies are analyzed according to the noise spectrum, and the initial denoising signal is dynamically optimized and adjusted according to the different frequency-domain characteristics of each sliced signal, so that each sliced signal can be optimally denoised. Finally, the optimized denoising parameters are used to denoise and reconstruct each sliced signal to obtain the denoised signal of the final target vibration signal. By fusing multiple task modules in this embodiment, the feature extraction of the sliced vibration signal is realized. According to the different frequency-domain characteristics of each sliced signal, the denoising parameters are dynamically adjusted, the denoising performance is adaptively optimized, the denoising ability for complex signals is enhanced, and the accuracy and efficiency of vibration signal denoising are improved.
[0030] In some embodiments of the present invention, as Figure 2 shown, Figure 2 is a schematic flowchart of an embodiment of step S101 provided by the present invention, including: Figure 1 S201. Build a denoising model based on DnCNN. The input of the denoising model is the noisy vibration signal, and the output is the denoised signal; Among them, based on the multi-task learning framework, a denoising model is built according to the DnCNN network and multi-scale feature extraction. On the basis of traditional denoising, this model adds a feature extraction task. In this embodiment, the choice of the denoising method is not limited to the DnCNN network, and the sliced vibration signal can also be denoised by wavelet transform, filter or other deep learning algorithms. Preferably, the DnCNN network is adopted.
[0031] S202. Based on the sliding window method, slice the target vibration signal to obtain a plurality of continuous sliced signals; It should be noted that in this embodiment, preferably, through the signal processing software, the sliding window method is used to slice the vibration signal to generate a plurality of sliced vibration signals, and the window size and sliding step of each sliced vibration signal are the same. S203. Obtain the initial denoised sliced signal according to the denoising model and the sliced signal; S204. Based on the multi-scale feature extraction algorithm, extract the frequency-domain characteristics of the sliced signal and the initial denoised sliced signal.
[0032] In this embodiment, when extracting features, the method is not limited to the multi-scale feature extraction algorithm, and the fast Fourier transform FFT can also be used to extract the frequency domain features of the slice vibration signal before and after denoising.
[0033] This embodiment can refine a complex vibration signal into multiple small slice signals by slicing the vibration signal according to a sliding window, can more accurately locate and analyze the characteristics of the vibration signal, and can denoise and extract features from the slices, providing an accurate basis for optimizing denoising parameters and improving the accuracy of denoising.
[0034] In some embodiments of the present invention, Figure 3 As shown, Figure 3 The present invention provides Figure 2 The flowchart of an embodiment of step S201 in FIG. 1 includes: S301, taking the noisy vibration signal as input and the denoised signal as output, constructing an initial model, wherein the initial model includes a denoising task and a feature extraction task; S302, constraining the output of the initial model according to the physical rules of the vibration signal to obtain a constrained model; Specifically, the denoising process in the vibration signal denoising model is physically constrained by the physical rule constraint module to ensure that the denoised signal conforms to the actual physical properties of the vibration signal, such as continuity and stability.
[0035] S303, constructing a loss function according to the denoising task and the feature extraction task; Specifically, the denoising model uses the mean square error loss function during the iterative training of the denoising task, and the formula of its loss function is: , in, is the denoising loss value, N is the number of vibration signal training samples, is the denoised output of the i-th vibration signal training sample, is the i-th vibration signal training sample.
[0036] Furthermore, when extracting features from the vibration signals before and after denoising, the model is also iteratively trained, and the mean square error loss function is also used in the training process. The formula of the loss function is: , in, is the feature extraction loss value, N is the number of vibration signal training samples, is the frequency domain feature of the denoised vibration signal training sample, is the frequency domain feature of the i-th vibration signal training sample before denoising.
[0037] Further, the denoising model combines the denoising task and the feature extraction task to ensure consistency in denoising and signal frequency domain conversion. During the combination process, the two tasks are weighted according to their task weights to obtain the total loss function. The denoising model for multiple tasks is iteratively trained using this total loss function. The calculation formula for the total loss function is: , where is the denoising loss value, is the feature extraction loss value, is the total loss value of the model, is a constant used to adjust the weights of the denoising task and the feature extraction task.
[0038] S304. Based on the reinforcement learning method and the loss function, iteratively optimize and adjust the constraint model to obtain the denoising model.
[0039] It should be noted that the structure and hyperparameters of the model are optimized through reinforcement learning, and the number of network layers, the size of the convolutional kernel, and the activation function are dynamically adjusted to further improve the performance of the model.
[0040] In this embodiment, by introducing physical rules to constrain the output of the denoising model, over-smoothing or distortion during the denoising process is avoided. The model is optimized through reinforcement learning, further improving the accuracy and efficiency of model denoising. Through the weighted loss function of the denoising task and the feature extraction task, the balance of the multi-task model is ensured, and the accuracy of model denoising is further improved.
[0041] In some embodiments of the present invention, as Figure 4 shown, Figure 4 is a schematic flowchart of an embodiment of step S102 provided by the present invention. Step S102 includes: Figure 1 S401. Construct an optimization function for the denoising parameters of the denoising model according to the noise power spectral density; S402. Determine the noise spectrum of the slice signal according to the frequency domain characteristics of the slice signal and the initial denoised slice signal. Specifically, the calculation formula for the noise spectrum is: , where is the frequency domain characteristic of the slice signal at the th moment, is the frequency domain characteristic of the denoised slice signal, is the noise spectrum of the slice signal, is the window time length of the slice signal.
[0042] S403. Determine the noise power spectral density of the slice signal according to the noise spectrum; S404. Based on the denoising parameter optimization function, iteratively optimize the denoising parameters according to the noise power spectral density to obtain the optimal denoising parameters.
[0043] Specifically, calculate the power spectral density by analyzing the energy or power of each frequency component of the noise spectrum. Usually, first score the noise spectrum and perform appropriate normalization processing to obtain the noise power spectral density.
[0044] Furthermore, compare the frequency-domain characteristics of the slice signal before and after denoising to obtain the noise power spectral density of the slice signal. Evaluate the denoising effect according to the noise power spectral density, adjust the denoising parameters according to the evaluation result, perform denoising processing on the slice signal using the adjusted denoising parameters, calculate the noise power spectral density of the new denoising result, construct an optimization function with the noise power spectral density as the evaluation criterion, and continuously iterate and evaluate and adjust the denoising parameters in combination with an optimization iteration algorithm (such as gradient descent, etc.). When the improvement of the denoising effect is no longer significant after continuous multiple iterations, it can be considered that convergence has been achieved, and this set of denoising parameters is the optimal denoising parameters. At the same time, when adjusting the denoising parameters, it is necessary to ensure that the parameters change within a reasonable range to avoid signal distortion caused by extreme parameters.
[0045] In this embodiment, by analyzing the noise spectrum of the slice signal, calculating the noise power spectral density, and iteratively optimizing the denoising parameters according to the noise power spectral density, the denoising parameters with the best denoising effect are obtained, which can dynamically adjust the denoising parameters according to the different frequency characteristics of each slice signal and can accurately and efficiently filter complex vibration signals.
[0046] In some embodiments of the present invention, as Figure 5 shown, Figure 5 is a schematic flowchart of an embodiment of step S404 provided by the present invention. S404 includes: Figure 4 S501. Based on the optimization algorithm, adjust the denoising parameters of the denoising model according to the noise power spectral density; S502. After multiple iterations, when the index value output by the denoising parameter optimization function reaches the preset convergence condition, stop the iteration to obtain the optimal denoising parameters.
[0047] It should be noted that there is no limitation on the specific optimization algorithm here according to the optimization algorithm (such as gradient descent, genetic algorithm, particle swarm optimization, etc.). According to the noise power spectral density, adjust the denoising parameters of the denoising model. Signals with larger noise power spectral density values require stronger denoising processing, and signals with weaker noise power spectral density values require appropriate reduction of the denoising intensity.
[0048] In this embodiment, by optimizing the denoising parameters according to the noise power spectral density of the evaluated slice signals, the denoising parameters can be dynamically adjusted according to the different frequency characteristics of each slice signal, and accurate and efficient filtering of complex vibration signals can be performed.
[0049] In some embodiments of the present invention, as Figure 6 shown, Figure 6 is a schematic flowchart of an embodiment of step S103 provided by the present invention. Step S103 includes: Figure 1 In some embodiments of the present invention, as S601. Adjust the hyperparameters of the denoising model according to the optimal denoising parameters to obtain an optimal denoising model; S602. Use the original slice signal as the input of the optimal denoising model and output the optimal denoised slice signal; S603. Reconstruct multiple optimal denoised slice signals according to the time-domain information to obtain the denoised signal of the target vibration signal.
[0050] Specifically, according to different denoising methods, their denoising parameters are also different, including filter type, filter order, cut-off frequency, etc. The slice signals are denoised using the optimal denoising parameters, and the denoising parameters are dynamically adjusted according to the different frequency characteristics of each slice signal to ensure that each slice signal of the vibration signal can be denoised by the optimal denoising parameters, effectively retaining the useful features in the signal while accurately denoising.
[0051] Furthermore, the target vibration signal contains time-domain information. During the process of slicing, denoising, and feature extraction of the target vibration signal, its time-domain information will not be lost. After denoising using the optimal denoising parameters, according to the timing information of each slice denoised signal, multiple slice denoised signals are recombined in the time order of the original vibration signal to obtain the final denoised vibration signal.
[0052] In some embodiments of the present invention, the denoising model includes an input layer, a denoising layer, a multi-scale feature extraction layer, multiple hidden layers, and an output layer; The input layer is used to receive the input target vibration signal; The multi-scale feature extraction layer is used to capture the multi-scale information of the vibration signal through multiple convolutional layers and extract the frequency characteristics and time characteristics; The hidden layer is used to extract deep features from different levels; The output layer is used to output the denoised vibration signal.
[0053] It should be noted that each vibration data is normalized in the input layer to standardize the range of the vibration data, and in the output layer, an inverse normalization operation is performed to restore the output vibration data to the range of the original vibration signal.
[0054] To better implement the vibration signal denoising method in the embodiments of the present invention, correspondingly, based on the vibration signal denoising method, as Figure 7 shown, an embodiment of the present invention further provides a vibration signal denoising device 700, including: A feature extraction module 701, configured to determine an initial denoised slice signal of the slice signal, a frequency domain feature of the slice signal, and a frequency domain feature of the initial denoised slice signal according to a plurality of consecutive slice signals of the target vibration signal; A denoising parameter optimization module 702, configured to determine a noise spectrum of each slice signal according to the frequency domain feature of the slice signal and the frequency domain feature of the initial denoised slice signal, and iteratively optimize the denoising parameter according to the noise spectrum to obtain an optimal denoising parameter; A denoising module 703, configured to perform denoising reconstruction on the slice signal according to the optimal denoising parameter to obtain a denoised signal of the target vibration signal.
[0055] The vibration signal denoising device 700 provided in the above embodiment can implement the technical solutions described in the embodiments of the above vibration signal denoising method. For the specific implementation principles of the above modules or units, reference can be made to the corresponding content in the embodiments of the above vibration signal denoising method, which will not be elaborated here.
[0056] In the embodiments of the present invention, the vibration signal denoising device can be an independent server, or a server network or server cluster composed of servers. For example, the vibration signal denoising device described in the embodiments of the present invention includes, but is not limited to, a computer, a network host, a single network server, a set of multiple network servers, or a cloud server composed of multiple servers. Among them, the cloud server is composed of a large number of computers or network servers based on cloud computing.
[0057] The above has introduced the vibration signal denoising method and device provided by the present invention in detail. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.
Claims
1. A vibration signal denoising method, characterized in that: include: Determining, according to a plurality of continuous slice signals of the target vibration signal, an initial denoised slice signal of the slice signal, a frequency domain feature of the slice signal, and a frequency domain feature of the initial denoised slice signal; Determine the noise spectrum of each slice signal according to the frequency domain characteristics of the slice signal and the frequency domain characteristics of the initial denoised slice signal, and iteratively optimize the denoising parameters according to the noise spectrum to obtain the optimal denoising parameters; The slice signal is denoised and reconstructed according to the optimal denoising parameters to obtain a denoised signal of the target vibration signal.
2. The vibration signal denoising method according to claim 1, characterized in that: Determining, according to a plurality of continuous slice signals of the target vibration signal, an initial denoised slice signal of the slice signal, a frequency domain feature of the slice signal, and a frequency domain feature of the initial denoised slice signal, comprising: A denoising model is constructed based on DnCNN, wherein the input of the denoising model is a noisy vibration signal, and the output is a denoised signal; Based on the sliding window method, the target vibration signal is sliced to obtain multiple continuous slice signals; Obtaining an initial denoised slice signal according to the denoising model and the slice signal; Based on a multi-scale feature extraction algorithm, frequency domain features of the slice signal and the initial denoised slice signal are extracted.
3. The vibration signal denoising method according to claim 2, characterized in that: Build a denoising model, including: Taking the noisy vibration signal as input and the denoised signal as output, an initial model is constructed, wherein the initial model includes a denoising task and a feature extraction task; According to the physical rules of the vibration signal, the output of the initial model is constrained to obtain a constrained model; Constructing a loss function according to the denoising task and the feature extraction task; Based on the reinforcement learning method and the loss function, the constraint model is iteratively optimized and adjusted to obtain a denoising model.
4. The vibration signal denoising method according to claim 3, characterized in that: The loss function of the denoising model is: ， in, is the loss value of the denoising task, is the loss value of the feature extraction task, is the model loss function value, is a constant, which is used to balance the weights of the vibration signal denoising task and the feature extraction task.
5. The vibration signal denoising method according to claim 4, characterized in that: Determine the noise spectrum of each slice signal according to the frequency domain characteristics of the slice signal and the frequency domain characteristics of the initial denoised slice signal, iteratively optimize the denoising parameters according to the noise spectrum, and obtain the optimal denoising parameters, including: Constructing an optimization function of denoising parameters of the denoising model according to the noise power spectrum density; Determining a noise spectrum of the slice signal according to the frequency domain characteristics of the slice signal and the initial denoised slice signal; Determine the noise power spectral density of the slice signal according to the noise spectrum; Based on the denoising parameter optimization function, the denoising parameters are iteratively optimized according to the noise power spectrum density to obtain optimal denoising parameters.
6. The vibration signal denoising method according to claim 5, characterized in that: The noise spectrum is calculated as: ， in, For the The frequency domain characteristics of the time-slice signal, is the frequency domain feature of the denoised slice signal, is the noise spectrum of the slice signal, is the window time length for slicing the signal.
7. The vibration signal denoising method according to claim 5, characterized in that: Based on the denoising parameter optimization function, the denoising parameters are iteratively optimized according to the noise power spectrum density to obtain the optimal denoising parameters, including: Based on the optimization algorithm, adjusting the denoising parameters of the denoising model according to the noise power spectrum density; After multiple iterations, when the index value output by the denoising parameter optimization function reaches a preset convergence condition, the iteration is stopped to obtain the optimal denoising parameter.
8. The vibration signal denoising method according to claim 7, characterized in that: De-noising and reconstructing the slice signal according to the optimal de-noising parameter to obtain a de-noised signal of the target vibration signal, including: Adjusting the hyperparameters of the denoising model according to the optimal denoising parameters to obtain an optimal denoising model; The original slice signal is used as the input of the optimal denoising model, and the optimal denoising slice signal is output; The plurality of optimal denoised slice signals are reconstructed according to time domain information to obtain a denoised signal of the target vibration signal.
9. The vibration signal denoising method according to claim 3, characterized in that: The denoising model includes an input layer, a denoising layer, a multi-scale feature extraction layer, a plurality of hidden layers and an output layer; The input layer is used to receive an input target vibration signal; The multi-scale feature extraction layer is used to capture multi-scale information of the vibration signal through multiple convolutional layers, and extract frequency characteristics and time characteristics; The hidden layer is used to extract deep features from different levels; The output layer is used to output the denoised vibration signal.
10. A vibration signal denoising device, characterized in that: include: A feature extraction module, used to determine, based on a plurality of continuous slice signals of the target vibration signal, an initial denoised slice signal of the slice signal, a frequency domain feature of the slice signal, and a frequency domain feature of the initial denoised slice signal; A denoising parameter optimization module, used to determine the noise spectrum of each slice signal according to the frequency domain characteristics of the slice signal and the frequency domain characteristics of the initial denoised slice signal, and iteratively optimize the denoising parameters according to the noise spectrum to obtain the optimal denoising parameters; The denoising module is used to denoise and reconstruct the slice signal according to the optimal denoising parameters to obtain a denoised signal of the target vibration signal.