Pressure-bearing equipment abnormal sound signal noise reduction and detection method, system, equipment and medium
Through the combination of adversarial training of adversarial generation and noise reduction models and one-dimensional convolutional neural network, the problem of poor noise processing effect of pressure-bearing equipment is solved, and high-quality signals are generated and accurately detected.
Patent Information
- Application Number
- CN202411189765.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-12-06
AI Technical Summary
In the prior art, the noise reduction method of the noise noise reduction method of the pressure-bearing equipment is insufficient in stability and adaptability, resulting in unsatisfactory detection effect and inability to accurately reflect the noise situation and operating status of the equipment.
Adversarial generation and noise reduction models are adopted, and the adversarial training of generator-encoder, decoder and discriminator is combined with one-dimensional convolutional neural network to reduce the noise and process the abnormal sound signals to generate high-quality detection signals.
It improves the detection accuracy and reliability of the abnormal sound signal of the pressure-bearing equipment, can effectively reduce noise, and accurately reflect the abnormal sound condition and operating status of the equipment.
Smart Images

Figure CN119091920A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of abnormal sound signal generation, noise reduction and abnormal sound detection of pressure-bearing equipment, and in particular to a method, system, equipment and medium for noise reduction and detection of abnormal sound signals of pressure-bearing equipment. Background Art
[0002] As the core equipment of the medium transportation and storage system, pressure equipment is widely used in industrial fields such as electricity and chemical industry. Failure of pressure equipment will seriously affect the safe and stable operation of the entire system. During operation, pressure equipment radiates sound information outward due to structural vibration, which is mixed with background noise to form abnormal sound signals. Once the pressure equipment is damaged or even fails, the abnormal sound signals of the pressure equipment will change accordingly. Therefore, timely detection of abnormal sound signals of pressure equipment can timely discover potential damage and failures of pressure equipment, thereby reducing operation and maintenance costs and improving the safety and reliability of the system, which has important theoretical value and practical significance. In addition, in actual engineering, the detection abnormal sound signals of pressure equipment are often mixed with background noises such as phase regulators, steam turbines, fans, pumps, cooling towers, etc., resulting in unsatisfactory application effects of existing detection methods, and directly applying them to the detection of abnormal sound conditions and operating status will result in low detection accuracy and unreliable detection results. Therefore, how to effectively reduce the noise of detection abnormal sound signals, generate high-quality detection abnormal sound signals, and achieve reliable and accurate detection of abnormal sound conditions and operating status of pressure equipment is worthy of in-depth research.
[0003] At present, signal denoising methods mainly include traditional signal separation and reconstruction and denoising based on deep learning models. Among them, the commonly used signal separation and reconstruction denoising methods mainly include empirical mode decomposition, variational mode decomposition, wavelet packet decomposition, etc. Signal denoising methods based on signal separation and reconstruction and their improved models usually decompose according to the characteristics of the signal, and combine experience for manual selection and reconstruction, resulting in insufficient stability and reliability of the denoising effect. On the other hand, with the development of deep learning models, signal denoising methods using deep learning models such as denoising autoencoders, generative adversarial networks, and transfer learning have emerged. These methods can achieve good denoising effects in specific data sets through training and testing of a large amount of data, but the training stability is insufficient and the adaptability is poor. Therefore, there is an urgent need for a detection method that can reduce the noise of the detection abnormal sound signal according to the characteristics of the detection abnormal sound signal, generate high-quality detection abnormal sound signals, and reliably and accurately reflect the abnormal sound situation and operating status of the pressure-bearing equipment. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides a method, system, device and medium for noise reduction and detection of abnormal sound signals of pressure-bearing equipment. The present invention overcomes the difficulty of artificial selection and reconstruction of the existing signal separation and reconstruction method, and the lack of stability and adaptability of the noise reduction method based on the deep learning model, which leads to the inability to accurately detect the abnormal sound situation and operating status of the pressure-bearing equipment, and effectively realizes the confrontational generation, noise reduction and abnormal sound detection of the abnormal sound signal of the pressure-bearing equipment.
[0005] To achieve the above object, the present invention adopts the following technical solutions:
[0006] A method for reducing and detecting abnormal sound signals of pressure-bearing equipment comprises the following steps:
[0007] Collect the noise-free abnormal sound signal of the pressure-bearing equipment in a noise-free environment and the noise-containing abnormal sound signal of the pressure-bearing equipment in a noise-containing environment;
[0008] Standardizing the noise-free and noise-containing signals, and dividing the standardized noise-free and noise-containing signals into training samples and test samples;
[0009] Inputting the training samples into the adversarial generation and denoising model for adversarial training, inputting the test samples into the adversarial generation and denoising model for testing, and obtaining a trained adversarial generation and denoising model;
[0010] Collect the noisy abnormal sound signals of pressure-bearing equipment under different pressure conditions in a noisy environment and input them into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction;
[0011] Performing standardization processing on the noise-reduced actually detected abnormal sound signal, and dividing the standardized noise-reduced actually detected abnormal sound signal into actual training samples and actual test samples;
[0012] The actual training samples and the actual test samples are input into a one-dimensional convolutional neural network to obtain abnormal sound conditions and operating status detection results of the pressure-bearing equipment.
[0013] To optimize the above technical solutions, the specific measures taken also include:
[0014] Further, the adversarial generation and denoising model includes a generative adversarial network, a denoising autoencoder and a deep residual network, the generative adversarial network includes a generator and a discriminator, the denoising autoencoder includes an encoder and a decoder, the generator and the encoder are combined to form a generator-encoder, and the deep residual network connects the convolutional layers of the generator-encoder and the decoder;
[0015] The generator-encoder includes 3 convolutional layers, the filter size of the convolutional layer of the generator-encoder is 3*1, the number is 128, and the number of channels is 128;
[0016] The decoder includes 2 deconvolution-convolution layers and 2 convolution layers connected in sequence, the filter size of the deconvolution-convolution layer and the first convolution layer of the decoder is 3*1, the number is 128, and the number of channels is 128, and the filter size of the second convolution layer of the decoder is 1*1, the number is 128, and the number of channels is 1;
[0017] The discriminator consists of 1 convolutional layer and 3 fully connected layers. The filter size of the convolutional layer of the discriminator is 3*1, the number of filters is 128, and the number of channels is 128. The size of the first fully connected layer of the discriminator is 1460*150, the size of the second fully connected layer is 150*150, and the size of the third fully connected layer is 1*2.
[0018] Furthermore, the adversarial training is specifically:
[0019] The training samples including the noise-free and noise-containing noise signals after the standardization processing are input into the generator-encoder for encoding to obtain the encoded noisy noise signal and the encoded noise-free noise signal; the encoded noisy noise signal is input into the decoder to obtain the noise-reduced noise signal; the noise-reduced noise signal is input into the generator-encoder to obtain the encoded noise-reduced noise signal; the encoded noisy noise signal, the encoded noise-free noise signal, and the encoded noise-reduced noise signal are input into the discriminator to obtain the authenticity judgment probability, the authenticity judgment probability is used to calculate the generator-encoder loss result and the discriminator loss result, the standardized noise-free noise signal and the noise-reduced noise signal are used to calculate the decoder loss result, the generator-encoder loss result is fed back to the generator-encoder, the decoder loss result is fed back to the decoder, and the discriminator loss result is fed back to the discriminator to complete the adversarial training.
[0020] Furthermore, the specific calculation method of the generator-encoder loss result is:
[0021] The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is:
[0022] L1=log(1-D(G(XN)))
[0023] Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal;
[0024] The mean square error of the noise reduction signal is calculated based on the deviation between the noise reduction signal and the noise-free signal after standardization. The expression is:
[0025]
[0026] Among them, L2 represents the mean square error of the noise reduction signal, XW i represents the i-th amplitude of the noise-free signal after normalization, XD i represents the i-th amplitude of the noise reduction signal, and n represents the length of the noise-free signal and the length of the noise reduction signal after the normalization process;
[0027] The generator-encoder loss result is obtained by superimposing the cross entropy of the encoded noisy noise signal and the mean square error of the denoised noise signal:
[0028] L E-G =L1+L2
[0029] Where, L E-G represents the generator-encoder loss result;
[0030] The decoder loss result is calculated by the mean square error of the noise reduction signal, and the expression is:
[0031] L H =L2
[0032] Among them, L H Indicates the decoder loss result.
[0033] Furthermore, the specific calculation method of the discriminator loss result is:
[0034] According to the authenticity judgment probability of the discriminator, the cross entropy of the coded noise-free foreign sound signal and the cross entropy of the coded noise-reduced foreign sound signal are calculated, and the expression is:
[0035] L3=log(1-D(G(XW)))
[0036] L4=log(1-D(G(XD)))
[0037] Wherein, L3 represents the cross entropy of the encoded noise-free noise signal, L4 represents the cross entropy of the encoded noise-reduced noise signal, XW represents the noise-free noise signal, and XD represents the noise-reduced noise signal; G(XW) represents the encoded noise-free noise signal, G(XD) represents the encoded noise-reduced noise signal, D(G(XW)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-free noise signal; D(G(XD)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-reduced noise signal;
[0038] The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is:
[0039] L1=log(1-D(G(XN)))
[0040] Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal;
[0041] The discriminator loss result is calculated by superimposing the cross entropy of the encoded noisy foreign sound signal, the cross entropy of the encoded noise-free foreign sound signal, and the cross entropy of the encoded denoised foreign sound signal, and the expression is:
[0042] L D =L1+L3+L4
[0043] Among them, L D Represents the discriminator loss result.
[0044] Furthermore, the generation of the abnormal sound signal actually detected after noise reduction is specifically as follows:
[0045] The noisy abnormal sound signal of the pressure-bearing equipment under different pressure conditions in a noisy environment is first input into the generator-encoder to obtain the encoded actual detection abnormal sound signal, and then the encoded actual detection abnormal sound signal is input into the decoder to obtain the actual detection abnormal sound signal after noise reduction.
[0046] Furthermore, the one-dimensional convolutional neural network includes 2 convolutional layers, 2 pooling layers, 1 fully connected layer and 1 Softmax layer. The filter size of the convolutional layer of the one-dimensional convolutional neural network is 6*1, and the number of channels is 8. The filter size of the pooling layer is 6*1, the step size is 6, and the number of channels is 8. The output number of the Softmax layer is 3; the one-dimensional convolutional neural network is constructed in the order of convolutional layer, pooling layer, convolutional layer, pooling layer, fully connected layer and Softmax layer.
[0047] The present invention also proposes a pressure equipment abnormal sound signal noise reduction and detection system, comprising:
[0048] Acoustic sensors are used to collect noise-free abnormal sound signals of pressure-bearing equipment in a noise-free environment and noise-containing abnormal sound signals of pressure-bearing equipment in a noise-containing environment, as well as to collect noise-containing abnormal sound signals of pressure-bearing equipment under different pressure conditions in a noise-containing environment;
[0049] The first data set processing module is used to perform standardization processing on the noise-free noise signal and the noisy noise signal, and divide the noise-free noise signal and the noisy noise signal after the standardization processing into training samples and test samples;
[0050] A model training module, used to input the training samples into the adversarial generation and denoising model for adversarial training, and input the test samples into the adversarial generation and denoising model for testing, so as to obtain a trained adversarial generation and denoising model;
[0051] The noise reduction module is used to input the noisy abnormal sound signals of the pressure-bearing equipment under different pressure conditions in the noise environment into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction;
[0052] The second data set processing module is used to perform standardization processing on the actual detected abnormal sound signal after noise reduction, and divide the standardized actual detected abnormal sound signal after noise reduction into actual training samples and actual test samples;
[0053] The pressure-bearing equipment detection module is used to input the actual training samples and the actual test samples into a one-dimensional convolutional neural network to obtain pressure-bearing equipment detection results.
[0054] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for reducing and detecting abnormal sound signals of pressure-bearing equipment as described above is implemented.
[0055] The present invention also proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the above-mentioned method for reducing and detecting abnormal sound signals of pressure-bearing equipment.
[0056] The beneficial effects of the present invention are:
[0057] 1) The present invention provides a method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment. The method uses noisy and noise-free abnormal sound signals to train adversarial generation and noise reduction models, and applies the trained adversarial generation and noise reduction models to the actual collected detection abnormal sound signals for noise reduction processing to generate high-quality detection abnormal sound signals. Finally, a one-dimensional convolutional neural network is used to detect the abnormal sound situation and operating status of the pressure-bearing equipment. This method avoids the shortcomings of strong background noise of the actual detection abnormal sound signal and the inability to accurately detect the abnormal sound situation and operating status of the pressure-bearing equipment by directly applying the actual detection abnormal sound signal. The accuracy of the detection of abnormal sound situation and operating status of the pressure-bearing equipment is improved by noise reduction and generation of high-quality detection abnormal sound signals.
[0058] 2) The adversarial generation and noise reduction model provided by the present invention fully utilizes the characteristics of the generator-encoder, decoder and discriminator to carry out adversarial training, which can perform noise reduction processing on the detected abnormal sound signal, thereby generating high-quality detected abnormal sound signal and improving the quality and reliability of the detected abnormal sound signal.
[0059] 3) The adversarial training method provided by the present invention calculates the authenticity judgment probability of the input noisy, noise-free and noise-reduced abnormal sound signals through the discriminator to construct the losses of the generator-encoder, decoder and discriminator, and uses the discriminator to improve the training speed and effect of the generator-encoder and decoder, and improve the noise reduction quality of the abnormal sound signal, thereby providing data support for the detection of abnormal sound conditions and operating status of pressure-bearing equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 It is a flowchart of a method for reducing and detecting abnormal sound signals of pressure-bearing equipment proposed in an embodiment of the present invention;
[0061] Figure 2 It is the structure of the adversarial generation and denoising model in the embodiment of the present invention.
[0062] Figure 3 It is a flow chart of detecting abnormal sound signal processing and adversarial training in an embodiment of the present invention.
[0063] Figure 4 It is a comparison diagram of the noise reduction effect of detecting abnormal sound signals in an embodiment of the present invention.
[0064] Figure 5 It is a statistical diagram of the SNR of the noise reduction signal obtained by different methods using the signal-to-noise ratio as the evaluation index. DETAILED DESCRIPTION
[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0066] Embodiment 1
[0067] The present invention proposes a method for reducing and detecting abnormal sound signals of pressure-bearing equipment. Figure 1 As shown, the following steps are included:
[0068] Step 1: Collect the noise-free abnormal sound signal of the pressure-bearing equipment in a noise-free environment and the noise-containing abnormal sound signal of the pressure-bearing equipment in a noise-containing environment;
[0069] Step 2: Standardize the noise-free and noise-containing signals, and divide the standardized noise-free and noise-containing signals into training samples and test samples;
[0070] Step 3: input the training samples into the adversarial generation and denoising model for adversarial training, input the test samples into the adversarial generation and denoising model for testing, and obtain a trained adversarial generation and denoising model;
[0071] Step 4: Collect the noisy abnormal sound signals of the pressure-bearing equipment under different pressure conditions in the noise environment and input them into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction;
[0072] Step 5: performing standardization processing on the noise-reduced actual detected abnormal sound signal, and dividing the standardized noise-reduced actual detected abnormal sound signal into actual training samples and actual test samples;
[0073] Step 6: Input the actual training samples and the actual test samples into a one-dimensional convolutional neural network to obtain abnormal sound conditions and operating status detection results of the pressure-bearing equipment.
[0074] The sampling time of the abnormal sound signal is 5s, the length of the abnormal sound signal is 1460, and the training sample and the test sample are randomly divided according to a ratio of 7:3. The training sample and the test sample may have certain differences due to the randomness of the division, but it does not affect the subsequent processing.
[0075] In step 1 and step 4, an acoustic sensor is used to collect abnormal sound signals. The acoustic sensor is 1 m above the middle side of the pressure-bearing equipment and is directed downward. The acoustic sensor is facing the pressure-bearing equipment.
[0076] It should be noted that the middle of the pressure-bearing equipment refers to the center of the length of the pressure-bearing equipment to be tested, and the 1m above refers to the highest point on the upper surface of the pressure-bearing equipment.
[0077] The standardization processing in step 2 and step 5 is a maximum minimization processing based on the maximum and minimum values of a certain sample. The expression of the maximum minimization processing is:
[0078]
[0079] Among them, x s (t) represents the signal after maximum-minimization processing, x(t) represents the original signal of a sample, min(x) and max(x) represent the minimum and maximum values of the original signal of a sample, respectively.
[0080] like Figure 2As shown, the adversarial generation and denoising model includes a generative adversarial network (GAN), a denoising autoencoder (DAE) and a deep residual network (ResNet), the generative adversarial network includes a generator and a discriminator, the denoising autoencoder includes an encoder and a decoder, the generator and the encoder are combined to form a generator-encoder, and the deep residual network connects the convolutional layers of the generator-encoder and the decoder; it should be noted that the ResNet connection adds the output results of the generator-encoder and the decoder.
[0081] The generator-encoder includes 3 convolutional layers, the filter size of the convolutional layer of the generator-encoder is 3*1, the number is 128, and the number of channels is 128;
[0082] The decoder includes 2 deconvolution-convolution layers and 2 convolution layers connected in sequence, the filter size of the deconvolution-convolution layer and the first convolution layer of the decoder is 3*1, the number is 128, and the number of channels is 128, and the filter size of the second convolution layer of the decoder is 1*1, the number is 128, and the number of channels is 1;
[0083] The discriminator includes 1 convolutional layer and 3 fully connected layers. The filter size of the convolutional layer of the discriminator is 3*1, the number is 128, and the number of channels is 128. The size of the first fully connected layer of the discriminator is 1460*150, the size of the second fully connected layer is 150*150, and the size of the third fully connected layer is 1*2. It should be noted that the parameters of the convolutional layer are the same in the generator-encoder, decoder and discriminator unless otherwise specified.
[0084] like Figure 3 As shown, the adversarial training is specifically:
[0085] The training samples including the noise-free and noise-containing noise signals after the normalization are input into the generator-encoder for encoding to obtain the encoded noisy noise signals and the encoded noise-free noise signals; the encoded noisy noise signals are input into the decoder to obtain the noise-reduced noise signals; the noise-reduced noise signals are input into the generator-encoder to obtain the encoded noise-reduced noise signals; the encoded noisy noise signals, the encoded noise-free noise signals, and the encoded noise-reduced noise signals are input into the discriminator to obtain the probability of true and false judgment, and the generator-encoder loss result and the discriminator loss result are calculated using the true and false judgment probability, and the decoder loss result is calculated using the normalized noise-free and noise-reduced noise signals, and the generator-encoder loss result is fed back to the generator-encoder, the decoder loss result is fed back to the decoder, and the discriminator loss result is fed back to the discriminator to complete the adversarial training. It should be noted that the probability of true and false judgment obtained by the discriminator is the probability that the discriminator judges that the input data is real data or false data.
[0086] The specific calculation method of the generator-encoder loss result is:
[0087] The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is:
[0088] L1=log(1-D(G(XN)))
[0089] Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal;
[0090] The mean square error of the noise reduction signal is calculated based on the deviation between the noise reduction signal and the noise-free signal after standardization. The expression is:
[0091]
[0092] Among them, L2 represents the mean square error of the noise reduction signal, XW i represents the i-th amplitude of the noise-free signal after normalization, XD i represents the i-th amplitude of the noise reduction signal, and n represents the length of the noise-free signal and the length of the noise reduction signal after the normalization process;
[0093] The generator-encoder loss result is obtained by superimposing the cross entropy of the encoded noisy noise signal and the mean square error of the denoised noise signal:
[0094] L E-G =L1+L2
[0095] Where, L E-Grepresents the generator-encoder loss result;
[0096] The decoder loss result is calculated by the mean square error of the noise reduction signal, and the expression is:
[0097] L H =L2
[0098] Among them, L H Indicates the decoder loss result.
[0099] The specific calculation method of the discriminator loss result is:
[0100] According to the authenticity judgment probability of the discriminator, the cross entropy of the coded noise-free foreign sound signal and the cross entropy of the coded noise-reduced foreign sound signal are calculated, and the expression is:
[0101] L3=log(1-D(G(XW)))
[0102] L4=log(1-D(G(XD)))
[0103] Wherein, L3 represents the cross entropy of the encoded noise-free noise signal, L4 represents the cross entropy of the encoded noise-reduced noise signal, XW represents the noise-free noise signal, and XD represents the noise-reduced noise signal; G(XW) represents the encoded noise-free noise signal, G(XD) represents the encoded noise-reduced noise signal, D(G(XW)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-free noise signal; D(G(XD)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-reduced noise signal;
[0104] The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is:
[0105] L1=log(1-D(G(XN)))
[0106] Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal;
[0107] The discriminator loss result is calculated by superimposing the cross entropy of the encoded noisy foreign sound signal, the cross entropy of the encoded noise-free foreign sound signal, and the cross entropy of the encoded denoised foreign sound signal, and the expression is:
[0108] L D =L1+L3+L4
[0109] Among them, L D Represents the discriminator loss result.
[0110] The actual detection of abnormal sound signals after generating noise reduction is specifically as follows:
[0111] The noisy abnormal sound signal of the pressure-bearing equipment under different pressure conditions in a noisy environment is first input into the generator-encoder to obtain the encoded actual detection abnormal sound signal, and then the encoded actual detection abnormal sound signal is input into the decoder to obtain the actual detection abnormal sound signal after noise reduction.
[0112] It should be noted that the actual detected abnormal sound signal in the process of generating the actual detected abnormal sound signal after noise reduction and the detected abnormal sound signal used in the adversarial training come from the same pressure-bearing device.
[0113] The one-dimensional convolutional neural network includes 2 convolutional layers, 2 pooling layers, 1 fully connected layer and 1 Softmax layer. The filter size of the convolutional layer of the one-dimensional convolutional neural network is 6*1, the number of channels is 8, the filter size of the pooling layer is 6*1, the step size is 6, the number of channels is 8, and the number of outputs of the Softmax layer is 3. The one-dimensional convolutional neural network is constructed in the order of convolutional layer, pooling layer, convolutional layer, pooling layer, fully connected layer and Softmax layer.
[0114] Embodiment 2
[0115] The present invention proposes a pressure equipment abnormal sound signal noise reduction and detection system corresponding to the method of embodiment 1, comprising:
[0116] Acoustic sensors are used to collect noise-free abnormal sound signals of pressure-bearing equipment in a noise-free environment and noise-containing abnormal sound signals of pressure-bearing equipment in a noise-containing environment, as well as to collect noise-containing abnormal sound signals of pressure-bearing equipment under different pressure conditions in a noise-containing environment;
[0117] The first data set processing module is used to perform standardization processing on the noise-free noise signal and the noisy noise signal, and divide the noise-free noise signal and the noisy noise signal after the standardization processing into training samples and test samples;
[0118] A model training module, used to input the training samples into the adversarial generation and denoising model for adversarial training, and input the test samples into the adversarial generation and denoising model for testing, so as to obtain a trained adversarial generation and denoising model;
[0119] The noise reduction module is used to input the noisy abnormal sound signals of the pressure-bearing equipment under different pressure conditions in the noise environment into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction;
[0120] The second data set processing module is used to perform standardization processing on the actual detected abnormal sound signal after noise reduction, and divide the standardized actual detected abnormal sound signal after noise reduction into actual training samples and actual test samples;
[0121] The pressure-bearing equipment detection module is used to input the actual training samples and the actual test samples into a one-dimensional convolutional neural network to obtain pressure-bearing equipment detection results.
[0122] The implementation methods of each module and module function in the system are completely consistent with the steps of the method in Example 1, so they will not be repeated here.
[0123] Embodiment 3
[0124] The present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for reducing and detecting abnormal sound signals of pressure-bearing equipment as described in Example 1 is implemented.
[0125] Embodiment 4
[0126] The present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the method for reducing and detecting abnormal sound signals of pressure-bearing equipment as described in the first embodiment.
[0127] In order to further verify the effectiveness of the present invention, the scheme of the present application is further described using a public pipeline leakage sound dataset:
[0128] The pipeline leakage sound dataset has three pressures of 0.2MPa, 0.4MPa and 0.5MPa. There are two environments, normal and noisy, under each pressure. There are 57 samples in each working condition. There are 2×3×57 types of pressure pipeline leakage detection abnormal sound signals with a length of 1460 (2 noise environments, 3 pressures, 57 sets of data for each working condition). The adversarial generation and denoising model of the present invention is trained using 0.2MPa noise-free and noisy abnormal sound signals, and applied to 0.4MPa noisy abnormal sound signals. The results are as follows: Figure 4 As shown. It can be seen that the amplitude of the noisy abnormal sound signal is higher and more chaotic. After the adversarial generation and noise reduction model processing, the amplitude of the noise reduction abnormal sound signal is reduced and is closer to the noise-free abnormal sound signal. The pressure pipeline leakage detection results are obtained by using the proposed method, as shown in Table 1. It can be seen that the leakage identification accuracy of the noisy abnormal sound signal is only 76.47%. The noise reduction abnormal sound signal invented can achieve an average accuracy of 91.17%, which is about 15% higher than the noisy abnormal sound signal and is close to the noise-free abnormal sound signal, verifying the accuracy and reliability of the method of this application.
[0129] Table 1 Leakage detection results of pressure pipelines
[0130]
[0131] In addition, it is compared with unsupervised denoising methods, including wavelet packet decomposition (WPD) denoising, empirical mode decomposition (EMD) denoising and denoising autoencoder (DAE) denoising. The SNR statistics of the denoised signals obtained by different methods are calculated using the signal-noise ratio (SNR) as the evaluation index. The results are shown in Figure 2. Figure 5 Compared with other methods, the SNR value distribution of the present invention is basically at the highest position, and the maximum SNR value exceeds 3.3dB, which is better than other methods, verifying the noise reduction effect of the present invention method.
[0132] In addition, the accuracy of pressure pipeline leakage detection of different noise reduction methods is calculated, as shown in Table 2. It can be seen that the accuracy of pressure pipeline leakage detection is improved by reducing the noise of noisy abnormal sound signals through different methods. The noise reduction methods based on WPD, EMD and DAE can improve the accuracy by 6%-10%, but the method of the present invention can achieve an accuracy of 91.17%, which is close to the detection effect of noise-free abnormal sound signals and is better than other methods. This shows that the present invention improves the accuracy of pipeline leakage detection by reducing the noise in the abnormal sound signal, further verifying the effectiveness and reliability of the method of the present application.
[0133] Table 2 Leakage detection results of pressure pipelines with different noise reduction methods
[0134] Sample Type Accuracy Noisy signal 76.47% Noise-free signal 94.12% WPD noise reduction and abnormal sound signal 82.35% EMD noise reduction and abnormal sound signal 80.31% DAE noise reduction and noise signal 86.27% Denoising abnormal sound signal of the present invention 91.17%
[0135] Compared with the prior art, the method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment of the present invention uses noisy and noise-free abnormal sound signals to train adversarial generation and noise reduction models, reduces the noise of abnormal sound detection signals through the trained adversarial generation and noise reduction models, generates high-quality abnormal sound detection signals, and combines with a one-dimensional convolutional neural network to realize the detection of abnormal sound conditions and operating status of pressure-bearing equipment. This avoids the shortcomings of strong background noise of actual abnormal sound detection signals and the inability to accurately detect abnormal sound conditions and operating status of pressure-bearing equipment by directly applying actual abnormal sound detection signals. The accuracy of abnormal sound condition and operating status detection of pressure-bearing equipment is improved by noise reduction and generation of high-quality abnormal sound detection signals.
[0136] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.
[0137] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0138] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment, characterized in that: The following steps are involved: Collect the noise-free abnormal sound signal of the pressure-bearing equipment in a noise-free environment and the noise-containing abnormal sound signal of the pressure-bearing equipment in a noise-containing environment; Standardizing the noise-free and noise-containing signals, and dividing the standardized noise-free and noise-containing signals into training samples and test samples; Inputting the training samples into the adversarial generation and denoising model for adversarial training, inputting the test samples into the adversarial generation and denoising model for testing, and obtaining a trained adversarial generation and denoising model; Collect the noisy abnormal sound signals of pressure-bearing equipment under different pressure conditions in a noisy environment and input them into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction; Performing standardization processing on the noise-reduced actually detected abnormal sound signal, and dividing the standardized noise-reduced actually detected abnormal sound signal into actual training samples and actual test samples; The actual training samples and the actual test samples are input into a one-dimensional convolutional neural network to obtain abnormal sound conditions and operating status detection results of the pressure-bearing equipment.
2. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 1, characterized in that: The adversarial generation and denoising model includes a generative adversarial network, a denoising autoencoder and a deep residual network, the generative adversarial network includes a generator and a discriminator, the denoising autoencoder includes an encoder and a decoder, the generator and the encoder are combined to form a generator-encoder, and the deep residual network connects the convolutional layers of the generator-encoder and the decoder; The generator-encoder includes 3 convolutional layers, the filter size of the convolutional layer of the generator-encoder is 3*1, the number is 128, and the number of channels is 128; The decoder includes 2 deconvolution-convolution layers and 2 convolution layers connected in sequence, the filter size of the deconvolution-convolution layer and the first convolution layer of the decoder is 3*1, the number is 128, and the number of channels is 128, and the filter size of the second convolution layer of the decoder is 1*1, the number is 128, and the number of channels is 1; The discriminator consists of 1 convolutional layer and 3 fully connected layers. The filter size of the convolutional layer of the discriminator is 3*1, the number of filters is 128, and the number of channels is 128. The size of the first fully connected layer of the discriminator is 1460*150, the size of the second fully connected layer is 150*150, and the size of the third fully connected layer is 1*2.
3. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 2, characterized in that: The adversarial training is specifically as follows: The training samples including the noise-free and noise-containing noise signals after the standardization processing are input into the generator-encoder for encoding to obtain the encoded noisy noise signal and the encoded noise-free noise signal; the encoded noisy noise signal is input into the decoder to obtain the noise-reduced noise signal; the noise-reduced noise signal is input into the generator-encoder to obtain the encoded noise-reduced noise signal; the encoded noisy noise signal, the encoded noise-free noise signal, and the encoded noise-reduced noise signal are input into the discriminator to obtain the authenticity judgment probability, the authenticity judgment probability is used to calculate the generator-encoder loss result and the discriminator loss result, the standardized noise-free noise signal and the noise-reduced noise signal are used to calculate the decoder loss result, the generator-encoder loss result is fed back to the generator-encoder, the decoder loss result is fed back to the decoder, and the discriminator loss result is fed back to the discriminator to complete the adversarial training.
4. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 3, characterized in that: The specific calculation method of the generator-encoder loss result is: The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is: L1=log(1-D(G(XN))) Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal; The mean square error of the noise reduction signal is calculated based on the deviation between the noise reduction signal and the noise-free signal after standardization. The expression is: Among them, L2 represents the mean square error of the noise reduction signal, XW i represents the i-th amplitude of the noise-free signal after normalization, XD i represents the i-th amplitude of the noise reduction signal, and n represents the length of the noise-free signal and the length of the noise reduction signal after the normalization process; The generator-encoder loss result is obtained by superimposing the cross entropy of the encoded noisy noise signal and the mean square error of the denoised noise signal: L E-G =L1+L2 Where, L E-G represents the generator-encoder loss result; The decoder loss result is calculated by the mean square error of the noise reduction signal, and the expression is: L H =L2 Among them, L H Indicates the decoder loss result.
5. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 3, characterized in that: The specific calculation method of the discriminator loss result is: According to the authenticity judgment probability of the discriminator, the cross entropy of the coded noise-free foreign sound signal and the cross entropy of the coded noise-reduced foreign sound signal are calculated, and the expression is: L3=log(1-D(G(XW))) L4=log(1-D(G(XD))) Wherein, L3 represents the cross entropy of the encoded noise-free noise signal, L4 represents the cross entropy of the encoded noise-reduced noise signal, XW represents the noise-free noise signal, and XD represents the noise-reduced noise signal; G(XW) represents the encoded noise-free noise signal, G(XD) represents the encoded noise-reduced noise signal, D(G(XW)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-free noise signal; D(G(XD)) represents the true and false judgment probability obtained when the discriminator identifies the encoded noise-reduced noise signal; The cross entropy of the encoded noisy foreign sound signal is calculated using the true and false judgment probability of the discriminator, and the expression is: L1=log(1-D(G(XN))) Where, L1 represents the cross entropy of the encoded noisy foreign sound signal, XN represents the noisy foreign sound signal, G(XN) represents the encoded noisy foreign sound signal, and D(G(XN)) represents the true and false judgment probability obtained by the discriminator when identifying the encoded noisy foreign sound signal; The discriminator loss result is calculated by superimposing the cross entropy of the encoded noisy foreign sound signal, the cross entropy of the encoded noise-free foreign sound signal, and the cross entropy of the encoded denoised foreign sound signal, and the expression is: <h2 style=";text-align:left;direction:ltr">L<h2 style=";text-align:left;direction:ltr"> D <h2 style=";text-align:left;direction:ltr"> (L1+L3+L4) Among them, L D Represents the discriminator loss result.
6. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 2, characterized in that: The actual detection of abnormal sound signals after generating noise reduction is specifically as follows: The noisy abnormal sound signal of the pressure-bearing equipment under different pressure conditions in a noisy environment is first input into the generator-encoder to obtain the encoded actual detection abnormal sound signal, and then the encoded actual detection abnormal sound signal is input into the decoder to obtain the actual detection abnormal sound signal after noise reduction.
7. The method for noise reduction and detection of abnormal sound signals of pressure-bearing equipment according to claim 1, characterized in that: The one-dimensional convolutional neural network includes 2 convolutional layers, 2 pooling layers, 1 fully connected layer and 1 Softmax layer. The filter size of the convolutional layer of the one-dimensional convolutional neural network is 6*1, and the number of channels is 8. The filter size of the pooling layer is 6*1, the step size is 6, and the number of channels is 8. The output number of the Softmax layer is 3; the one-dimensional convolutional neural network is constructed in the order of convolutional layer, pooling layer, convolutional layer, pooling layer, fully connected layer and Softmax layer.
8. A system for reducing and detecting abnormal sound signals of pressure-bearing equipment, characterized in that: include: Acoustic sensors are used to collect noise-free abnormal sound signals of pressure-bearing equipment in a noise-free environment and noise-containing abnormal sound signals of pressure-bearing equipment in a noise-containing environment, as well as to collect noise-containing abnormal sound signals of pressure-bearing equipment under different pressure conditions in a noise-containing environment; The first data set processing module is used to perform standardization processing on the noise-free noise signal and the noisy noise signal, and divide the noise-free noise signal and the noisy noise signal after the standardization processing into training samples and test samples; A model training module, used to input the training samples into the adversarial generation and denoising model for adversarial training, and input the test samples into the adversarial generation and denoising model for testing, so as to obtain a trained adversarial generation and denoising model; The noise reduction module is used to input the noisy abnormal sound signals of the pressure-bearing equipment under different pressure conditions in the noise environment into the trained adversarial generation and noise reduction model to generate the actual detection abnormal sound signals after noise reduction; The second data set processing module is used to perform standardization processing on the actual detected abnormal sound signal after noise reduction, and divide the standardized actual detected abnormal sound signal after noise reduction into actual training samples and actual test samples; The pressure-bearing equipment detection module is used to input the actual training samples and the actual test samples into a one-dimensional convolutional neural network to obtain pressure-bearing equipment detection results.
9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for reducing and detecting abnormal sound signals of pressure-bearing equipment as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the method for reducing and detecting abnormal sound signals of pressure-bearing equipment as described in any one of claims 1 to 7.
