Acoustic emission effective signal extraction method and device, equipment and storage medium

Abstract

The invention discloses an acoustic emission effective signal extraction method, device and equipment and a storage medium, and the method comprises the steps: collecting an initial acoustic emission signal, and carrying out the frequency band decomposition of the initial acoustic emission signal through wavelet packet transformation, and obtaining a plurality of frequency band waveforms; inputting each frequency band waveform into a preset auto-encoder model for reconstruction feature filtering to obtain an initial filtering signal; the initial filtering signal is input into a preset feature extraction model for time sequence feature filtering, an acoustic emission effective signal is obtained, and the preset feature extraction model is used for conducting time sequence feature analysis on the initial filtering signal. Compared with the prior art, the method has the advantages that the initial acoustic emission signals are subjected to frequency band decomposition by adopting wavelet packet transformation, then different frequency band waveforms are input into the preset auto-encoder model for reconstruction feature filtering, and then the filtered signals are subjected to time sequence feature analysis, so that background noise is effectively eliminated, and the accuracy of the acoustic emission signals is improved. The extraction of the sound emission effective signal in the strong background noise environment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of civil engineering structure monitoring, and in particular to a method, device, equipment and storage medium for extracting effective acoustic emission signals. Background Art

[0002] Civil engineering structures subjected to dynamic loads are prone to fatigue failure, such as highway and railway steel bridges, offshore oil platforms, wind power towers, steel crane beams, and industrial equipment supports. Among these building structures, the fatigue problem of welded steel structures is the most significant. Since fatigue failure often exhibits extremely brittle fracture characteristics, it is extremely sudden and often accompanied by catastrophic engineering accidents. In order to ensure the safety of building structures and avoid catastrophic accidents, it is of great significance to monitor fatigue damage of welded steel structures that are prone to fatigue failure. Acoustic emission signals are a type of stress wave generated by the evolution of internal defects in materials. Compared with traditional non-destructive testing, acoustic emission methods have the characteristics of dynamic and real-time, and are less restricted by the geometric shape of the test piece or structure. Since the acoustic emission signal originates from the damage evolution of the structure or material, the signal itself carries rich damage evolution information. Although the acoustic emission technology has produced fruitful research results in the laboratory environment since it was proposed, it is often restricted in its promotion and application in actual engineering. The main reason is that the actual engineering environment is often accompanied by strong background noise, which often makes the effective acoustic emission signal (the signal representing the evolution of structural or material damage) difficult to identify or even completely annihilated.

[0003] Most of the existing noise reduction methods for acoustic emission signals introduce thresholds based on signal decomposition, or determine the thresholds through experience, or determine the thresholds through artificial intelligence algorithms, and then remove the noise part, extract the effective part and reconstruct the signal. These methods are generally more applicable when the frequency domain difference between noise and effective signals is large, or when the intensity of noise signals in some frequency bands is smaller than that of effective acoustic emission signals. However, in the field of civil engineering steel structures, the intensity of background noise is usually greater than the effective signal released by material fatigue damage, and the frequency band of noise is relatively wide, which is generally difficult to distinguish in the frequency domain.

[0004] Therefore, there is an urgent need for a method for extracting effective acoustic emission signals, which can solve the technical problem in the prior art that effective acoustic emission signals are difficult to be identified under the interference of strong background noise. Summary of the invention

[0005] The main purpose of the present invention is to provide a method, device, equipment and storage medium for extracting effective acoustic emission signals, aiming to solve the technical problem in the prior art that effective acoustic emission signals are difficult to be identified under the interference of strong background noise.

[0006] To achieve the above object, the present invention provides a method for extracting effective acoustic emission signals, the method comprising the following steps:

[0007] Collecting an initial acoustic emission signal, and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms;

[0008] Inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal;

[0009] The initial filtered signal is input into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal. The preset feature extraction model is used to perform time series feature analysis on the initial filtered signal.

[0010] Optionally, the preset feature extraction model includes a layer of CNN, four layers of GRU and a layer of ANN, the layer of CNN is used to compress data and extract local features of data, the four layers of GRU are used to learn temporal features, and the layer of ANN is used for signal classification.

[0011] Optionally, before the step of collecting the initial acoustic emission signal and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms, the step further includes:

[0012] Acquire a pure noise signal sample library, perform a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, wherein the number of the first target waveforms is equal to the number of the frequency band waveforms;

[0013] Establishing an autoencoder model based on each of the first target waveforms, wherein the autoencoder model comprises a plurality of autoencoder modules, and the number of the autoencoder modules is equal to the number of the first target waveforms;

[0014] The autoencoder model is trained using the first target waveform to obtain a preset autoencoder model, and a corresponding loss threshold is determined based on the preset autoencoder model.

[0015] Optionally, before the step of collecting the initial acoustic emission signal and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms, the step further includes:

[0016] Acquire a pure fatigue damage acoustic emission signal sample library, perform the target wavelet packet tree transform on the pure fatigue damage acoustic emission signal sample library, decompose the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into a plurality of second target waveforms, wherein the number of the second target waveforms is equal to the number of the first target waveforms;

[0017] Establishing a feature extraction model based on each of the second target waveforms, wherein the feature extraction model comprises a plurality of feature extraction modules, the number of the feature extraction modules is equal to the number of the second target waveforms, and the feature extraction modules comprise a layer of CNN, four layers of GRU and a layer of ANN;

[0018] The feature extraction model is trained using the second target waveform to obtain a preset feature extraction model.

[0019] Optionally, the step of inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal includes:

[0020] Inputting each of the frequency band waveforms into a preset autoencoder model to obtain a reconstructed waveform;

[0021] Determining the loss value of each of the reconstructed waveforms and the corresponding each of the frequency band waveforms;

[0022] The reconstructed waveforms corresponding to the loss values ​​less than or equal to the loss threshold are set to zero, and the reconstructed waveforms corresponding to the loss values ​​greater than the loss threshold are added to obtain an initial filtered signal.

[0023] Optionally, the step of inputting the initial filtered signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal includes:

[0024] Inputting the initial filtered signal into a preset feature extraction model to perform a time series feature analysis on the initial filtered signal through the preset feature extraction model to obtain a time series feature analysis result;

[0025] Classifying the initial filtered signal based on the time series feature analysis result to obtain a noise signal to be removed and a fatigue damage signal;

[0026] The noise signal to be eliminated in the initial filtered signal is eliminated, and the fatigue damage signal is used as an effective acoustic emission signal.

[0027] Optionally, before the step of acquiring a pure noise signal sample library and performing a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, the step further includes:

[0028] Determine whether the steel structure in the service environment is a steel structure whose establishment time does not exceed a preset time threshold;

[0029] If the steel structure is of a construction time that does not exceed a preset time threshold, the noise signal in the service environment is collected, and a pure noise signal sample library is generated based on the noise signal in the service environment;

[0030] If the steel structure has a construction time exceeding a preset time threshold, detecting whether fatigue damage occurs to the steel structure;

[0031] If the steel structure does not suffer from fatigue damage, the noise signal under the service environment is collected, and a pure noise signal sample library is generated based on the noise signal under the service environment.

[0032] In addition, to achieve the above-mentioned purpose, the present invention also proposes a device for extracting effective acoustic emission signals, the device comprising:

[0033] A signal acquisition module is used to collect an initial acoustic emission signal and perform frequency band decomposition on the initial acoustic emission signal through wavelet packet transform to obtain multiple frequency band waveforms;

[0034] A first filtering module, used for inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal;

[0035] The second filtering module is used to input the initial filtering signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal. The preset feature extraction model is used to perform time series feature analysis on the initial filtering signal.

[0036] In addition, to achieve the above-mentioned purpose, the present invention also proposes an acoustic emission effective signal extraction device, which includes: a memory, a processor, and an acoustic emission effective signal extraction program stored in the memory and executable on the processor, wherein the acoustic emission effective signal extraction program is configured to implement the steps of the acoustic emission effective signal extraction method as described above.

[0037] In addition, to achieve the above objectives, the present invention also proposes a storage medium, on which is stored an acoustic emission effective signal extraction program, which, when executed by a processor, implements the steps of the acoustic emission effective signal extraction method as described above.

[0038] The present invention discloses collecting an initial acoustic emission signal, and performing frequency band decomposition on the initial acoustic emission signal through wavelet packet transform to obtain multiple frequency band waveforms; inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal; inputting the initial filtered signal into a preset feature extraction model for time series feature filtering to obtain an acoustic emission effective signal, and the preset feature extraction model is used to perform time series feature analysis on the initial filtered signal. Since the present invention first uses wavelet packet transform to perform frequency band decomposition on the initial acoustic emission signal, and then inputs different frequency band waveforms into a preset autoencoder model for reconstructing feature filtering, and then performs time series feature filtering on the filtered signal using a preset feature extraction model, secondary filtering of the initial acoustic emission signal is achieved. Compared with the prior art, the present invention effectively eliminates background noise and achieves the extraction of effective acoustic emission signals in a strong background noise environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the first embodiment of the method for extracting effective acoustic emission signals of the present invention;

[0040] Figure 2 This is a typical signal diagram of fatigue crack growth in welding nodes;

[0041] Figure 3 This is a typical noise signal diagram in an actual engineering environment;

[0042] Figure 4 It is a schematic diagram of the first filtering effect in the method for extracting effective acoustic emission signals of the present invention;

[0043] Figure 5 It is a schematic diagram of the effect of extracting effective acoustic emission signals in the effective acoustic emission signal extraction method of the present invention;

[0044] Figure 6 It is a flow chart of a second embodiment of the method for extracting effective acoustic emission signals of the present invention;

[0045] Figure 7 It is a schematic diagram of the acoustic emission signal processing flow in the acoustic emission effective signal extraction method of the present invention;

[0046] Figure 8 It is a schematic flow chart of a third embodiment of the method for extracting effective acoustic emission signals of the present invention;

[0047] Fig. 9 It is a structural block diagram of the first embodiment of the acoustic emission effective signal extraction device of the present invention;

[0048] Fig.10 It is a structural diagram of an acoustic emission effective signal extraction device in a hardware operating environment involved in an embodiment of the present invention.

[0049] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0051] The embodiment of the present invention provides a method for extracting effective acoustic emission signals, referring to Figure 1 , Figure 1 It is a flow chart of the first embodiment of the method for extracting effective acoustic emission signals of the present invention.

[0052] In this embodiment, the acoustic emission effective signal extraction method includes the following steps:

[0053] Step S10: collecting an initial acoustic emission signal, and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms.

[0054] It should be noted that the execution subject of this embodiment can be a computer server device with data processing, network communication and program running functions, such as a server, a tablet computer, a personal computer, etc., or an electronic device capable of realizing the above functions, an acoustic emission effective signal extraction device, etc. The following takes the acoustic emission effective signal extraction device as an example to illustrate this embodiment and the following embodiments.

[0055] It should be understood that the above initial acoustic emission signal may be an acoustic emission signal of mixed noise and fatigue damage in an actual service environment, that is, an acoustic emission signal of mixed noise and fatigue damage in an actual engineering environment.

[0056] It is understandable that wavelet packet transform is an extension of wavelet transform. Traditional wavelet transform mainly decomposes the low-frequency part of the signal and ignores the high-frequency part. Wavelet packet transform decomposes not only the low-frequency part, but also the high-frequency part, that is, it divides all frequency bands more finely. This transformation is calculated through a series of basis functions, which include four parameters: frequency, displacement, expansion and inversion, which can describe the local characteristics of the signal at different frequencies.

[0057] It should be noted that, according to the frequency range of the signal to be analyzed, a suitable wavelet packet basis function can be selected, and then the optimal wavelet packet tree is selected according to the selected wavelet packet basis function, and then the initial acoustic emission signal is decomposed by the wavelet packet transform of the optimal wavelet packet tree.

[0058] Step S20: Input each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal.

[0059] It should be noted that the preset autoencoder model is an unsupervised learning neural network model that works by learning the compressed representation of the input data (encoding) and reconstructing the original input from this representation (decoding). The preset autoencoder model mainly consists of two parts: encoder and decoder. The encoder is responsible for mapping the input data to a low-dimensional latent space representation, with the aim of extracting the effective features of the input data and compressing it into a more compact form. The decoder is responsible for mapping the output of the encoder (i.e., the latent space representation) back to the original input space, with the aim of reconstructing the original input data as accurately as possible.

[0060] In a specific implementation, each of the frequency band waveforms can be input into a preset autoencoder model to obtain a reconstructed waveform; then the loss value of each of the reconstructed waveforms and the corresponding frequency band waveforms is determined; finally, the reconstructed waveform corresponding to the loss value less than or equal to the loss threshold of the preset autoencoder model is set to zero, and the reconstructed waveforms corresponding to the loss value greater than the loss threshold of the preset autoencoder model are added to obtain an initial filtered signal.

[0061] Step S30: Inputting the initial filtered signal into a preset feature extraction model to perform time series feature filtering to obtain an effective acoustic emission signal, wherein the preset feature extraction model is used to perform time series feature analysis on the initial filtered signal.

[0062] It should be explained that in actual engineering, the acoustic emission signals collected by the sensor can be divided into two types: one is the signal representing the environmental background noise, and the other is the effective acoustic emission signal representing the fatigue damage evolution of the welding node. These two signals are usually difficult to distinguish simply by frequency screening, that is, they are both broadband signals. However, the signal representing the evolution of fatigue damage has its own unique characteristics, which are reflected in both the time domain and the frequency, especially the time sequence in the time domain. Figure 2 , Figure 2 This is a typical signal diagram of fatigue crack growth in welding nodes. The horizontal axis represents time (unit: milliseconds ms) and the vertical axis represents signal strength (unit: millivolts mv). When fatigue cracks in steel structure welding nodes grow, the effective acoustic emission signal generated by energy release has typical bursts, presents a strong triangle in time sequence, and decays quickly. Figure 3 , Figure 3 This is a typical noise signal diagram in an actual engineering environment. The horizontal axis represents time (unit: milliseconds ms) and the vertical axis represents signal strength (unit: millivolts mv). Figure 2 and Figure 3 It can be seen that there is a great difference in timing between the effective acoustic emission signal and the noise signal. Therefore, this embodiment can utilize this difference in timing and use a preset feature extraction model to perform timing feature analysis on the initial filtering signal to achieve secondary filtering of the initial acoustic emission signal.

[0063] refer to Figure 4 , Figure 4 Schematic diagram of the first filtering effect in the acoustic emission effective signal extraction method of the present invention, in which the target signal represents the acoustic emission effective signal, the synthetic signal represents the collected initial acoustic emission signal including the background noise signal and the acoustic emission effective signal, and the signal after the first extraction represents the initial filtered signal obtained after the reconstruction feature filtering. Figure 4 It can be seen that reconstructing feature filtering based on the preset autoencoder model can filter out most of the noise signals. In order to further eliminate the background noise signal, this embodiment performs time series feature filtering through a preset feature extraction model to achieve secondary filtering of the initial acoustic emission signal, thereby further eliminating the background noise signal.

[0064] refer to Figure 5 , Figure 5 It is a schematic diagram of the acoustic emission effective signal extraction effect in the acoustic emission effective signal extraction method of the present invention. In the figure, SNR represents signal-to-noise ratio, target signal represents acoustic emission effective signal, synthetic signal represents the collected initial acoustic emission signal including background noise signal and acoustic emission effective signal, primary extraction represents reconstruction feature filtering based on a preset autoencoder model, and secondary extraction represents time series feature filtering based on a preset feature extraction model on the basis of primary extraction. Figure 4 and Figure 5 The horizontal axis represents time (unit: microsecond μs), and the vertical axis represents signal strength (unit: millivolt mv).

[0065] It needs to be explained that the above-mentioned preset feature extraction model includes a layer of CNN (Convolutional Neural Networks), four layers of GRU (Gated Recurrent Unit) and a layer of ANN (Artificial Neural Networks), wherein the layer of CNN is used to compress data and extract local features of data, the four layers of GRU are used to learn time series features, and the layer of ANN is used for signal classification.

[0066] In a specific implementation, the initial filtered signal can be input into a preset feature extraction model to perform a timing feature analysis on the initial filtered signal through the preset feature extraction model to obtain a timing feature analysis result; the initial filtered signal is classified based on the timing feature analysis result to obtain a noise signal to be eliminated and a fatigue damage signal; the noise signal to be eliminated in the initial filtered signal is eliminated, and the fatigue damage signal is used as an effective acoustic emission signal.

[0067] The present embodiment discloses collecting an initial acoustic emission signal, and performing frequency band decomposition of the initial acoustic emission signal through wavelet packet transform to obtain multiple frequency band waveforms; inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal; inputting the initial filtered signal into a preset feature extraction model for time series feature filtering to obtain an acoustic emission effective signal, and the preset feature extraction model is used to perform time series feature analysis on the initial filtered signal. Since the present embodiment first uses wavelet packet transform to perform frequency band decomposition of the initial acoustic emission signal, and then inputs different frequency band waveforms into a preset autoencoder model for reconstructing feature filtering, and then performs time series feature filtering on the filtered signal using a preset feature extraction model, secondary filtering of the initial acoustic emission signal is achieved. Compared with the prior art, the present embodiment effectively eliminates background noise and achieves the extraction of effective acoustic emission signals in a strong background noise environment.

[0068] refer to Figure 6 , Figure 6 It is a flow chart of the second embodiment of the method for extracting effective acoustic emission signals of the present invention.

[0069] Based on the first embodiment above, in this embodiment, before step S10, the following steps are further included:

[0070] Step S101: obtaining a pure noise signal sample library, performing a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, wherein the number of the first target waveforms is equal to the number of the frequency band waveforms.

[0071] It should be understood that the pure noise signal sample library may be a sample data set of acoustic emission signals in a steel structure service environment collected to ensure no fatigue damage. In addition, the pure noise signal sample library can generally cover most noise types in a service environment.

[0072] It should be explained that the target wavelet packet tree transform can be the optimal wavelet packet tree transform adjusted and optimized by comprehensively considering multiple factors such as signal characteristics, analysis objectives, wavelet packet basis selection and decomposition layer number.

[0073] It is understandable that when performing target wavelet packet tree transform on the noise signal in the pure noise signal sample library, a suitable wavelet packet basis should be selected. The user can make a selection based on the actual situation, for example, a wavelet packet basis between Db4 and Db10. The more layers of the target wavelet tree, the better. However, considering the computational complexity, the number of layers of the target wavelet tree in this embodiment can be user-defined, for example, set to 4 layers.

[0074] It should be noted that, through the decomposition of the target wavelet packet tree, the noise signal in the pure noise signal sample library is decomposed into n first target waveforms, where n is equal to the number of terminal nodes of the target wavelet packet tree.

[0075] Step S102: establishing an autoencoder model based on each of the first target waveforms, wherein the autoencoder model comprises a plurality of autoencoder modules, and the number of the autoencoder modules is equal to the number of the first target waveforms.

[0076] Step S103: training the autoencoder model using the first target waveform to obtain a preset autoencoder model, and determining a corresponding loss threshold based on the preset autoencoder model.

[0077] It should be noted that, for n first target waveforms, an autoencoder model is established, the autoencoder model includes n autoencoder modules, and then the autoencoder model is trained based on the first target waveform. When the waveform reconstruction effect is best, the corresponding autoencoder model is used as the preset autoencoder model, and then the loss threshold distribution of the preset autoencoder model is determined, and the 4 quantiles with larger losses can be selected as the loss threshold.

[0078] In a specific implementation, the step of inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal includes: inputting each of the frequency band waveforms into a preset autoencoder model to obtain a reconstructed waveform; determining the loss value of each of the reconstructed waveforms and the corresponding frequency band waveforms; setting the reconstructed waveform corresponding to a loss value less than or equal to the loss threshold to zero, and adding the reconstructed waveforms corresponding to the loss value greater than the loss threshold to obtain an initial filtered signal.

[0079] It should be understood that the reconstructed waveform corresponding to the loss value less than or equal to the loss threshold is a background noise signal, and the reconstructed waveform corresponding to the loss value greater than the loss threshold is a possible fatigue damage signal. Therefore, after the n waveforms pass through the n encoder modules of the preset autoencoder model, those determined to be noise signals are directly set to zero, and those determined to be possible fatigue damage signals are retained. Finally, the retained waveforms are added together to obtain the signal after the first filtering, that is, the initial filtered signal.

[0080] Furthermore, before step S10, the method further includes:

[0081] Step S104: obtaining a pure fatigue damage acoustic emission signal sample library, performing the target wavelet packet tree transform on the pure fatigue damage acoustic emission signal sample library, and decomposing the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into a plurality of second target waveforms, wherein the number of the second target waveforms is equal to the number of the first target waveforms.

[0082] It is understandable that the pure fatigue damage acoustic emission signal sample library may be a signal sample library formed by collecting noise-free pure fatigue damage acoustic emission signals in a laboratory environment.

[0083] Step S105: Establish a feature extraction model based on each of the second target waveforms, the feature extraction model includes multiple feature extraction modules, the number of the feature extraction modules is equal to the number of the second target waveforms, and the feature extraction modules include one layer of CNN, four layers of GRU and one layer of ANN.

[0084] It should be noted that the one-layer CNN is used to compress data and extract local features of data, the four-layer GRU is used to learn time series features, and the one-layer ANN is used for signal classification.

[0085] Step S106: training the feature extraction model using the second target waveform to obtain a preset feature extraction model.

[0086] It should be understood that training the feature extraction model by using the second target waveform can achieve adjustment and optimization of the model parameters of the feature extraction model, thereby obtaining a preset feature extraction model.

[0087] refer to Figure 7 , Figure 7 It is a schematic diagram of the acoustic emission signal processing flow in the acoustic emission effective signal extraction method of the present invention. The figure includes two filtering processes. The first filtering mainly uses the optimal wavelet packet tree transform + preset autoencoder model to perform signal reconstruction feature filtering on the acoustic emission signal of mixed noise and fatigue damage (i.e., initial acoustic emission signal) in the actual service environment. The second filtering is performed after the first filtering, mainly using the preset feature extraction model to perform signal timing feature filtering to obtain the acoustic emission signal containing the evolution of fatigue damage of the welding node (i.e., acoustic emission effective signal). It should be noted that before filtering the acoustic emission signal of mixed noise and fatigue damage in the actual service environment, two steps need to be performed. Step one: learn the background noise characteristics in the actual service environment; Step two: learn the timing characteristics of the fatigue signal of the welding node.

[0088] In step one, a large number of acoustic emission signals under the service environment of the steel structure need to be collected to obtain a pure noise signal sample library, and then the pure noise signal in the pure noise signal sample library is decomposed into an optimal wavelet packet tree to obtain a waveform corresponding to the end node of the optimal wavelet packet tree (i.e., a first target waveform), and then the first target waveform is input into an autoencoder model (Autoencoder model) for training to obtain a preset autoencoder model, and a corresponding loss threshold is determined based on the preset autoencoder model, the autoencoder model includes a plurality of autoencoder modules, the number of the autoencoder modules is equal to the number of the first target waveform, that is, the autoencoder modules correspond to the total number of end nodes of the optimal wavelet packet tree, and the end nodes of n optimal wavelet packet trees correspond to n Autoencoder modules (the autoencoder module includes an encoder and a decoder, the encoder is on the same side as the input layer, and the decoder is on the same side as the output layer).

[0089] In step 2, it is necessary to collect signals obtained through fatigue tests in a laboratory environment to obtain a pure fatigue damage acoustic emission signal sample library, and then perform optimal wavelet packet tree decomposition on the pure fatigue damage acoustic emission signal sample library to decompose the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into waveforms corresponding to the end nodes of the optimal wavelet packet tree (i.e., the second target waveform), and then input the second target waveform into a feature extraction model (the feature extraction model includes CNN+GRU+ANN) for training to obtain a preset feature extraction model, wherein CNN represents that Conv1d in the figure is a convolutional neural network.

[0090] In a specific implementation, after performing step 1 to learn the background noise characteristics in the actual service environment and step 2 to learn the timing characteristics of the fatigue signal of the welding node, the acoustic emission signals of the mixed noise and fatigue damage in the actual service environment are collected, and the initial acoustic emission signal is decomposed into frequency bands through optimal wavelet packet tree decomposition to obtain multiple frequency band waveforms, and then each of the frequency band waveforms is input into a preset autoencoder model to obtain a reconstructed waveform; the loss value of each of the reconstructed waveforms and the corresponding frequency band waveforms is determined; the reconstructed waveform corresponding to the loss value less than or equal to the loss threshold is set to zero, and the reconstructed waveform corresponding to the loss value greater than the loss threshold is reset to zero. The waveforms are added to obtain an initial filtered signal (i.e., a reconstructed signal); the initial filtered signal is then input into a preset feature extraction model to perform a time series feature analysis on the initial filtered signal through the preset feature extraction model to obtain a time series feature analysis result; the initial filtered signal is classified based on the time series feature analysis result to obtain a noise signal to be eliminated (noise) and a fatigue damage signal (damage signal); the noise signal to be eliminated in the initial filtered signal is eliminated, and the fatigue damage signal is used as an effective acoustic emission signal, and finally an effective acoustic emission signal (i.e., an acoustic emission signal containing the evolution of fatigue damage of the welding node) is output.

[0091] The present embodiment discloses obtaining a pure noise signal sample library, performing a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, wherein the number of the first target waveforms is equal to the number of the frequency band waveforms; establishing an autoencoder model based on each of the first target waveforms, wherein the autoencoder model includes a plurality of autoencoder modules, wherein the number of the autoencoder modules is equal to the number of the first target waveforms; training the autoencoder model through the first target waveform to obtain a preset autoencoder model, and determining a corresponding loss threshold based on the preset autoencoder model; obtaining a pure fatigue damage acoustic emission signal sample library, performing the target wavelet packet tree transform on the pure fatigue damage acoustic emission signal sample library to decompose the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into a plurality of second target waveforms, wherein the number of the second target waveforms is equal to the number of the first target waveforms; establishing a feature extraction model based on each of the second target waveforms, wherein the feature extraction model includes a plurality of feature extraction modules, wherein the number of the feature extraction modules is equal to the number of the second target waveforms, and the feature extraction module includes a layer of CNN, four layers of GRU and a layer of ANN; training the feature extraction model through the second target waveform to obtain a preset feature extraction model. Compared with the prior art, this embodiment realizes the training of feature extraction models and feature extraction models by learning the background noise characteristics in the actual service environment and learning the timing characteristics of the fatigue signal of the welding node, thereby improving the model's effect of eliminating background noise signals.

[0092] refer to Figure 8 , Figure 8 It is a flow chart of the third embodiment of the method for extracting effective acoustic emission signals of the present invention.

[0093] Based on the above embodiments, in this embodiment, before step S101, the following steps are further included:

[0094] Step S01: Determine whether the steel structure in the service environment is a steel structure whose establishment time does not exceed a preset time threshold.

[0095] Step S02: If the construction time of the steel structure does not exceed the preset time threshold, the noise signal in the service environment is collected, and a pure noise signal sample library is generated based on the noise signal in the service environment.

[0096] Step S03: If the steel structure is constructed for a time period exceeding a preset time threshold, then detecting whether fatigue damage occurs to the steel structure.

[0097] Step S04: If the steel structure does not suffer from fatigue damage, the noise signal under the service environment is collected, and a pure noise signal sample library is generated based on the noise signal under the service environment.

[0098] It is understandable that if it is a newly built steel structure (i.e., a steel structure in a service environment is a steel structure whose construction time does not exceed a preset time threshold), then the noise signal in its service environment can be directly collected. If conditions permit, the more data samples of the noise signal, the better. If it is a steel structure that has been in service for a period of time (i.e., a steel structure in a service environment is a steel structure whose construction time exceeds a preset time threshold), then it is necessary to first conduct corresponding inspections to determine whether the steel structure has fatigue damage. If not, it can be collected and constructed in the same way as a newly built steel structure. Pure noise signal sample library.

[0099] This embodiment discloses a method for judging whether a steel structure in a service environment is a steel structure whose establishment time does not exceed a preset time threshold; if it is a steel structure whose establishment time does not exceed a preset time threshold, the noise signal in the service environment is collected, and a pure noise signal sample library is generated based on the noise signal in the service environment; if it is a steel structure whose establishment time exceeds a preset time threshold, whether fatigue damage occurs in the steel structure is detected; if the steel structure does not have fatigue damage, the noise signal in the service environment is collected, and a pure noise signal sample library is generated based on the noise signal in the service environment. Compared with the prior art, this embodiment ensures that the pure noise signal sample library only contains pure noise signals of steel structures in a service environment, thereby improving the accuracy of subsequent model training and ensuring the reliability of effective acoustic emission signal extraction.

[0100] In addition, an embodiment of the present invention further provides a storage medium, on which is stored an acoustic emission effective signal extraction program, which, when executed by a processor, implements the steps of the acoustic emission effective signal extraction method as described above.

[0101] Reference Fig. 9 , Fig. 9 This is a structural block diagram of the first embodiment of the acoustic emission effective signal extraction device of the present invention.

[0102] like Fig. 9 As shown, the acoustic emission effective signal extraction device provided in the embodiment of the present invention includes: a signal acquisition module 901 , a first filtering module 902 and a second filtering module 903 .

[0103] The signal acquisition module 901 is used to acquire an initial acoustic emission signal and perform frequency band decomposition on the initial acoustic emission signal through wavelet packet transformation to obtain multiple frequency band waveforms.

[0104] The first filtering module 902 is used to input each of the frequency band waveforms into a preset autoencoder model to perform reconstruction feature filtering to obtain an initial filtered signal.

[0105] The second filtering module 903 is used to input the initial filtering signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal. The preset feature extraction model is used to perform time series feature analysis on the initial filtering signal.

[0106] The second filtering module 903 is also used to input the initial filtering signal into a preset feature extraction model, so as to perform a timing feature analysis on the initial filtering signal through the preset feature extraction model to obtain a timing feature analysis result; classify the initial filtering signal based on the timing feature analysis result to obtain a noise signal to be eliminated and a fatigue damage signal; eliminate the noise signal to be eliminated in the initial filtering signal, and use the fatigue damage signal as an effective acoustic emission signal.

[0107] The embodiment of the present device discloses collecting an initial acoustic emission signal, and performing frequency band decomposition of the initial acoustic emission signal through wavelet packet transform to obtain multiple frequency band waveforms; inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal; inputting the initial filtered signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal, and the preset feature extraction model is used to perform time series feature analysis on the initial filtered signal. Since the embodiment of the present device first uses wavelet packet transform to perform frequency band decomposition of the initial acoustic emission signal, and then inputs different frequency band waveforms into a preset autoencoder model for reconstructing feature filtering, and then performs time series feature filtering on the filtered signal using a preset feature extraction model, secondary filtering of the initial acoustic emission signal is achieved. Compared with the prior art, the embodiment of the present device effectively eliminates background noise and achieves the extraction of effective acoustic emission signals in a strong background noise environment.

[0108] Based on the first embodiment of the effective acoustic emission signal extraction device of the present invention, a second embodiment of the effective acoustic emission signal extraction device of the present invention is proposed.

[0109] In this embodiment, the signal acquisition module 901 is also used to obtain a pure noise signal sample library, perform a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into multiple first target waveforms, and the number of the first target waveforms is equal to the number of the frequency band waveforms; establish an autoencoder model based on each of the first target waveforms, the autoencoder model includes multiple autoencoder modules, and the number of the autoencoder modules is equal to the number of the first target waveforms; train the autoencoder model through the first target waveform to obtain a preset autoencoder model, and determine the corresponding loss threshold based on the preset autoencoder model.

[0110] The signal acquisition module 901 is also used to obtain a pure fatigue damage acoustic emission signal sample library, perform the target wavelet packet tree transform on the pure fatigue damage acoustic emission signal sample library, decompose the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into multiple second target waveforms, and the number of the second target waveforms is equal to the number of the first target waveforms; establish a feature extraction model based on each of the second target waveforms, the feature extraction model includes multiple feature extraction modules, the number of the feature extraction modules is equal to the number of the second target waveforms, and the feature extraction module includes a layer of CNN, four layers of GRU and one layer of ANN; train the feature extraction model through the second target waveform to obtain a preset feature extraction model.

[0111] Other embodiments or specific implementations of the acoustic emission effective signal extraction device of the present invention can refer to the above-mentioned method embodiments, which will not be described in detail here.

[0112] The present application provides an acoustic emission effective signal extraction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the acoustic emission effective signal extraction method in the above-mentioned embodiment 1.

[0113] Reference below Fig.10 , which shows a schematic diagram of the structure of an effective acoustic emission signal extraction device suitable for implementing the embodiment of the present application. The effective acoustic emission signal extraction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.10 The acoustic emission effective signal extraction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0114] like Fig.10As shown, the acoustic emission effective signal extraction device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the acoustic emission effective signal extraction device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the acoustic emission effective signal extraction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows an acoustic emission effective signal extraction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0115] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0116] The acoustic emission effective signal extraction device provided by the present application adopts the acoustic emission effective signal extraction method in the above embodiment, which can solve the technical problem in the prior art that the acoustic emission effective signal is difficult to be identified under the interference of strong background noise. Compared with the prior art, the beneficial effects of the acoustic emission effective signal extraction device provided by the present application are the same as the beneficial effects of the acoustic emission effective signal extraction method provided by the above embodiment, and the other technical features in the acoustic emission effective signal extraction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0117] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0118] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0119] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0120] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0121] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0122] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for extracting effective acoustic emission signals, characterized in that: The method includes: Collecting an initial acoustic emission signal, and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms; Inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal; The initial filtered signal is input into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal. The preset feature extraction model is used to perform time series feature analysis on the initial filtered signal.

2. The method according to claim 1, characterized in that The preset feature extraction model includes a layer of CNN, four layers of GRU and a layer of ANN, the layer of CNN is used to compress data and extract local features of data, the four layers of GRU are used to learn temporal features, and the layer of ANN is used for signal classification.

3. The method according to claim 1, characterized in that Before the step of collecting the initial acoustic emission signal and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms, the method further includes: Acquire a pure noise signal sample library, perform a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, wherein the number of the first target waveforms is equal to the number of the frequency band waveforms; Establishing an autoencoder model based on each of the first target waveforms, wherein the autoencoder model comprises a plurality of autoencoder modules, and the number of the autoencoder modules is equal to the number of the first target waveforms; The autoencoder model is trained using the first target waveform to obtain a preset autoencoder model, and a corresponding loss threshold is determined based on the preset autoencoder model.

4. The method according to claim 3, characterized in that Before the step of collecting the initial acoustic emission signal and performing frequency band decomposition on the initial acoustic emission signal by wavelet packet transform to obtain multiple frequency band waveforms, the method further includes: Acquire a pure fatigue damage acoustic emission signal sample library, perform the target wavelet packet tree transform on the pure fatigue damage acoustic emission signal sample library, decompose the pure fatigue damage acoustic emission signal in the pure fatigue damage acoustic emission signal sample library into a plurality of second target waveforms, wherein the number of the second target waveforms is equal to the number of the first target waveforms; Establishing a feature extraction model based on each of the second target waveforms, wherein the feature extraction model comprises a plurality of feature extraction modules, the number of the feature extraction modules is equal to the number of the second target waveforms, and the feature extraction modules comprise a layer of CNN, four layers of GRU and a layer of ANN; The feature extraction model is trained using the second target waveform to obtain a preset feature extraction model.

5. The method according to claim 3, characterized in that The step of inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal comprises: Inputting each of the frequency band waveforms into a preset autoencoder model to obtain a reconstructed waveform; Determining the loss value of each of the reconstructed waveforms and the corresponding each of the frequency band waveforms; The reconstructed waveforms corresponding to the loss values ​​less than or equal to the loss threshold are set to zero, and the reconstructed waveforms corresponding to the loss values ​​greater than the loss threshold are added to obtain an initial filtered signal.

6. The method according to claim 1, characterized in that The step of inputting the initial filtered signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal comprises: Inputting the initial filtered signal into a preset feature extraction model to perform a time series feature analysis on the initial filtered signal through the preset feature extraction model to obtain a time series feature analysis result; Classifying the initial filtered signal based on the time series feature analysis result to obtain a noise signal to be removed and a fatigue damage signal; The noise signal to be eliminated in the initial filtered signal is eliminated, and the fatigue damage signal is used as an effective acoustic emission signal.

7. The method according to claim 3, characterized in that Before the step of acquiring a pure noise signal sample library and performing a target wavelet packet tree transform on the pure noise signal sample library to decompose the noise signal in the pure noise signal sample library into a plurality of first target waveforms, the step further includes: Determine whether the steel structure in the service environment is a steel structure whose establishment time does not exceed a preset time threshold; If the steel structure is of a construction time that does not exceed a preset time threshold, the noise signal in the service environment is collected, and a pure noise signal sample library is generated based on the noise signal in the service environment; If the steel structure has a construction time exceeding a preset time threshold, detecting whether fatigue damage occurs to the steel structure; If the steel structure does not suffer from fatigue damage, the noise signal under the service environment is collected, and a pure noise signal sample library is generated based on the noise signal under the service environment.

8. An acoustic emission effective signal extraction device, characterized in that: The device comprises: A signal acquisition module is used to collect an initial acoustic emission signal and perform frequency band decomposition on the initial acoustic emission signal through wavelet packet transform to obtain multiple frequency band waveforms; A first filtering module, used for inputting each of the frequency band waveforms into a preset autoencoder model for reconstructing feature filtering to obtain an initial filtered signal; The second filtering module is used to input the initial filtering signal into a preset feature extraction model for time series feature filtering to obtain an effective acoustic emission signal. The preset feature extraction model is used to perform time series feature analysis on the initial filtering signal.

9. An acoustic emission effective signal extraction device, characterized in that: The device comprises: a memory, a processor, and an acoustic emission effective signal extraction program stored in the memory and executable on the processor, wherein the acoustic emission effective signal extraction program is configured to implement the steps of the acoustic emission effective signal extraction method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores an acoustic emission effective signal extraction program, and when the acoustic emission effective signal extraction program is executed by the processor, the steps of the acoustic emission effective signal extraction method according to any one of claims 1 to 7 are implemented.

Images