Rotating machine fault diagnosis method based on confrontation noise reduction convolutional neural network

Abstract

The invention relates to a rotating machine fault diagnosis method and device based on an adversarial noise reduction convolutional neural network. The method comprises the following steps: firstly, acquiring a vibration signal and preprocessing the vibration signal to obtain a preprocessed signal sample; then, constructing an adversarial noise reduction convolutional neural network model based on a BP neural network and a convolutional neural network; inputting the pre-processed signal sample and a real sample into the confrontation noise reduction convolutional neural network model for alternate training; and finally, pre-processing a to-be-diagnosed vibration signal, inputting the pre-processed vibration signal into the trained confrontation noise reduction convolutional neural network model, and outputting a rotating machinery fault diagnosis result. By combining a confrontation mechanism, a denoising technology and a deep learning framework, the fault diagnosis performance of the rotating machine is remarkably improved in the aspects of a strong noise environment, limited sample conditions and multi-working-condition adaptability, and a high-practicability innovative solution is provided for health monitoring and fault prediction of the rotating machine in the industrial field.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to a rotating machinery fault diagnosis method based on an adversarial denoising convolutional neural network. Background Art

[0002] As an important power equipment widely used in the industrial field, rotating machinery has a complex structure and is often in a high temperature, high speed and strong noise working environment. Therefore, it is of great significance to conduct health assessment on rotating machinery.

[0003] In the prior art, the condition monitoring and diagnosis system can make full use of the large amount of data generated by rotating machinery in the manufacturing process, and mine effective information to provide guidance for actual production and equipment maintenance, thereby avoiding the occurrence of catastrophic events. It is mainly composed of signal processing and pattern recognition algorithms. Among them, signal processing is very effective in feature extraction and selection, such as wavelet analysis, morphological filtering, stochastic resonance, blind source separation and other methods. However, these methods usually require sophisticated technology and rich expert experience. In addition, pattern recognition has also been widely used in mechanical fault diagnosis. For example, support vector machine (SVM), hidden Markov model (HMM), k nearest neighbor algorithm and deep learning models that have emerged in recent years.

[0004] In recent years, deep learning has provided a new research direction for condition monitoring and diagnosis systems. Specifically, many deep learning frameworks have been proposed in recent years and have achieved remarkable results, such as deep belief networks (DBNs), convolutional neural networks (CNNs), stacked denoising autoencoders (SDAEs), long short-term memory networks (LSTMs), and recurrent neural networks (RNNs). However, the above methods do not fully consider the following issues: (1) The robustness of the model. Industrial sites usually have uncertain interference such as strong noise, which places high demands on the robustness of the model. (2) The generalization ability of the model. In the industrial field, the number of labeled fault samples is limited, which can easily lead to overfitting. In addition, due to changes in mechanical working conditions, bearing quality, etc., there are usually distribution differences between test samples and training samples, which can significantly reduce the diagnostic performance.

[0005] In addition, according to the application of deep learning models in condition monitoring and diagnosis systems, the current training samples are mainly divided into two categories: (1) time domain. The fault features in the time domain are not significant and are easily affected by environmental noise. (2) frequency domain. The fault resonance band in the frequency domain is significant and is more suitable for training models than time domain samples. In order to reduce training time and difficulty, the front part of the spectrum is usually selected as the training sample, but it is easy to mistakenly delete important resonance bands. Although the full spectrum is a better choice, the increase in signal length will increase training time and difficulty. Conversely, reducing the signal length will reduce the frequency resolution and cause the loss of important details.

[0006] Therefore, in the related technology, there is an urgent need for a method that can accurately and efficiently identify rotating machinery faults in a strong noise environment and under limited sample conditions. Summary of the invention

[0007] Based on this, it is necessary to provide a rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network, which can accurately and efficiently identify rotating machinery faults in a strong noise environment and under limited sample conditions, in response to the above technical problems.

[0008] In a first aspect, the present application provides a method for fault diagnosis of rotating machinery based on an adversarial denoising convolutional neural network. The method comprises:

[0009] Acquire the vibration signal and perform preprocessing to obtain a preprocessed signal sample;

[0010] Construct an adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network;

[0011] Inputting the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training;

[0012] After preprocessing, the vibration signal to be diagnosed is input into the trained adversarial denoising convolutional neural network model to output the rotating machinery fault diagnosis result.

[0013] Optionally, in one embodiment of the present application, the preprocessing includes:

[0014] The vibration signal is processed by using a fast Fourier transform and a moving maximum method to obtain a processed spectrum.

[0015] Optionally, in one embodiment of the present application, the preprocessing further includes:

[0016] The processed spectrum is dynamically erased using an optimized erasing coefficient.

[0017] Optionally, in one embodiment of the present application, inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training includes:

[0018] Map the random vector into the BP neural network to generate false samples corresponding to the real samples;

[0019] The real samples, preprocessed signal samples and false samples are input into the convolutional neural network to determine the authenticity of the samples and classify the samples.

[0020] Optionally, in one embodiment of the present application, inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training further includes:

[0021] Fixing the parameters of the BP neural network and optimizing the loss function of the convolutional neural network discriminator;

[0022] The parameters of the convolutional neural network are fixed, and the loss function of the convolutional neural network generator is optimized.

[0023] Optionally, in one embodiment of the present application, the method further includes:

[0024] The t-SNE technique is used to visualize the features of the input layer, the first pooling layer, the second pooling layer, and the Softmax layer of the adversarial denoising convolutional neural network model.

[0025] In a second aspect, the present application also provides a rotating machinery fault diagnosis device based on an adversarial denoising convolutional neural network. The device comprises:

[0026] A data preprocessing module is used to obtain and preprocess the vibration signal to obtain a preprocessed signal sample;

[0027] Model building module, used to build an adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network;

[0028] A model training module, used for inputting the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training;

[0029] The fault diagnosis module is used to pre-process the vibration signal to be diagnosed, input it into the trained adversarial denoising convolutional neural network model, and output the rotating machinery fault diagnosis result.

[0030] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the steps of the methods described in the above embodiments.

[0031] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in each of the above embodiments are implemented.

[0032] The above-mentioned method for fault diagnosis of rotating machinery based on adversarial denoising convolutional neural network, first, obtains vibration signal and preprocesses it to obtain preprocessed signal sample; then, constructs adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network; then, inputs the preprocessed signal sample and real sample into the adversarial denoising convolutional neural network model for alternating training; finally, after preprocessing the vibration signal to be diagnosed, inputs it into the trained adversarial denoising convolutional neural network model, and outputs the fault diagnosis result of rotating machinery. That is to say, by introducing the moving maximum value method, the anti-noise ability of training samples is significantly enhanced, and the dynamic erasing technology is combined to simulate noise interference, which helps the model to more accurately capture the fault characteristics in the noise environment, thereby improving the diagnostic accuracy and robustness, and by using the adversarial mechanism to generate more representative training samples, expand the sample size, and enhance the generalization ability of the model, so that it can maintain high diagnostic accuracy under limited sample conditions. By combining adversarial mechanism, denoising technology and deep learning framework, the fault diagnosis performance of rotating machinery is significantly improved in strong noise environment, limited sample conditions and multi-working condition adaptability, providing a practical and innovative solution for health monitoring and fault prediction of rotating machinery in the industrial field. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 An application environment diagram of a rotating machinery fault diagnosis method based on an adversarial denoising convolutional neural network in one embodiment;

[0034] Figure 2 A schematic diagram of a flow chart of a rotating machinery fault diagnosis method based on an adversarial denoising convolutional neural network in one embodiment;

[0035] Figure 3 A schematic diagram of data preprocessing in one embodiment;

[0036] Figure 4 is a schematic diagram of different training samples in one embodiment;

[0037] Figure 5 is a schematic diagram of erasure probabilities under different erasure coefficients in one embodiment;

[0038] Figure 6 is a schematic diagram of diagnostic accuracy under different erasure coefficients in one embodiment;

[0039] Figure 7 A schematic diagram of an adversarial denoising convolutional neural network model in one embodiment;

[0040] Figure 8 A schematic diagram of feature visualization in one embodiment;

[0041] Fig. 9A schematic diagram of the training loss of a denoising convolutional neural network model in one embodiment;

[0042] Fig.10 It is a structural block diagram of a rotating machinery fault diagnosis device based on an adversarial denoising convolutional neural network in one embodiment;

[0043] Fig.11 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] The present invention provides a method for diagnosing rotating machinery faults based on an adversarial denoising convolutional neural network, which can be applied to Figure 1 In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be integrated on the server, or it can be placed on the cloud or other network servers. Among them, the terminal can be but not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0046] In one embodiment, Figure 2 As shown in the figure, a rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network is provided. Figure 1 The server in the example is used to illustrate the following steps:

[0047] S201: Acquire a vibration signal and perform preprocessing to obtain a preprocessed signal sample.

[0048] In the embodiment of the present application, first, a vibration signal is obtained. In a specific application, a bearing experimental data set provided by Case Western Reserve University (CWRU) is used, and the sampling frequency is 12kHz. The data set contains data of normal state and fault state, where the fault state includes inner ring fault (IRF), outer ring fault (ORF) and rolling element fault (BF), and each fault corresponds to four different damage sizes. In this application, nine fault types of the drive end (DE) at 1772rpm are studied as an example, including (a) inner race fault (IRF), damage size 0.007; (b) rolling element fault (BF), damage size 0.007; (c) outer race fault (ORF), damage size 0.007; (d) inner race fault (IRF), damage size 0.014; (e) rolling element fault (BF), damage size 0.014; (f) outer race fault (ORF), damage size 0.014; (g) inner race fault (IRF), damage size 0.21; (h) rolling element fault (BF), damage size 0.21; (i) outer race fault (ORF). The training and test samples are expanded by segmenting the original vibration signal and setting overlap, and each obtained sample contains 10240 data points. Generally, the characteristics of a signal in the frequency domain are more significant than those in the time domain, and the peaks in the spectrum can often better characterize the fault characteristics. In addition, in the industrial field, the sampling frequency of the signal is usually high, even reaching several megahertz. For the signal used to construct the training sample, a short signal will reduce the frequency resolution, resulting in insufficient utilization of the frequency information; while a long signal will increase the sample dimension and the complexity of the model training. Therefore, in order to solve this problem, the acquired vibration signal is preprocessed to obtain a preprocessed signal sample.

[0049] Specifically, in one embodiment of the present application, the preprocessing includes:

[0050] The vibration signal is processed by using a fast Fourier transform and a moving maximum method to obtain a processed spectrum.

[0051] In one embodiment of the present application, Figure 3 As shown, the spectrum of the segmented samples is obtained by fast Fourier transform (FFT), and then the spectrum is processed by the moving maximum method to obtain a training sample with a length of 1024. Assume that the original signal is x signal , whose length is l signal , l w is the length of the sliding window, and the step size is l s The i-th resampling point obtained by the moving maximum method It can be expressed as:

[0052]

[0053] Among them, p is the pth point of the original signal. By moving the maximum value method, we can get a length of l resample The resampled signal, where l resample =1+(l signal -l w ) / l s .

[0054] In one embodiment of the present application, in order to verify the effectiveness of the moving maximum method, two signals are analyzed: one is the original signal (OS), and the other is a noise signal (NS) generated by adding strong noise to the OS. The length of OS and NS is 2048. First, the signal is converted into the frequency domain by Fast Fourier Transform (FFT) to obtain a training sample with a length of 1024 (such as Figure 4 In addition, the OS and NS signals with a length of 10240 are first converted into training samples with a length of 5120 by FFT (as shown in Figure 4 (c)-(d)). Subsequently, the moving maximum method is applied to the obtained spectrum, where the step size is 5 and the sliding window length is 5, so that the training samples with a length of 1024 are obtained (as shown in Figure 4 (e)-(f)). Obviously, Figure 4 Compared with the sample in (b), Figure 4 The samples in (d) have better noise resistance, but high-dimensional samples will increase the complexity of model training. Figure 4 (d) Figure 4 The spectrum distribution in (f) is closer to the original signal and has a lower dimension, which shows that the moving maximum method can enhance the noise resistance of the sample without increasing the sample dimension. In addition, a data set from Case Western Reserve University (CWRU) (data set name is IR007) is selected to study the sample correlation coefficients obtained by different methods under different load conditions. The results are shown in Table 1. As can be seen from Table 1, the correlation coefficient in the frequency domain is significantly higher than the correlation coefficient in the time domain, and the moving maximum method has the highest correlation coefficient, which shows that the moving maximum method can reduce the difference in sample distribution under different working conditions to a certain extent.

[0055] Table 1. Correlation coefficients obtained from different load samples by different methods.

[0056]

[0057] In one embodiment of the present application, the preprocessing further includes:

[0058] The processed spectrum is dynamically erased using an optimized erasing coefficient.

[0059] In one embodiment of the present application, Figure 3 As shown in the figure, during the preprocessing process, the training samples are erased with random erasure probability in each traversal. The erasure point of the sample is directly set to zero to simulate noise interference. Then all samples are normalized to [-1,1] to obtain the final training samples. Its erasure probability (ep) is expressed as:

[0060]

[0061] Where p is a random variable that satisfies the uniform distribution U[0.2,1.0], and k is the erasure coefficient of (0,+∞).

[0062] The erasure probability (ep) corresponding to different erasure coefficient k values ​​is as follows: Figure 5 As shown in Figure 1, as k increases, the degree of sample damage increases dramatically, which means that more noise is added to the sample. When k→0 and ep=0, it means that the sample is not damaged and all information is retained; when k→∞ and ep=1, it means that all information of the sample is lost. It should be noted that the optimal value of k is different for different data sets. In specific applications, such as Figure 6 As shown in Figure 2, compared with the noise-free sample, when the erasure coefficient is too large, the characteristics of the noise sample are overly destroyed, resulting in low diagnostic accuracy. When the erasure coefficient is too small, the characteristics of the noise sample can be captured in the noise environment, significantly improving the diagnostic accuracy. The best results can be obtained.

[0063] S203: Construct an adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network.

[0064] In the present application embodiment, Figure 7 As shown in Figure 2, a convolutional neural network model for adversarial denoising is constructed based on BP neural network and convolutional neural network. BP neural network and CNN are used to play the roles of generator G and discriminator D respectively. BP neural network has a 3-layer structure, specifically 256-512-1024. The specific parameter configuration of CNN is shown in Table 2.

[0065] Table 2. Parameter settings of CNN.

[0066]

[0067] S205: Inputting the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training.

[0068] In the present application embodiment, Figure 7 As shown, the real samples generated by the BP neural network Preprocessing samples and false samples The samples are input into ADCNN as training samples, thus ensuring that the model learns the characteristics of both real samples and noisy samples at the same time, improving the robustness of the model.

[0069] Specifically, in one embodiment of the present application, inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training includes:

[0070] S301: Map the random vector into the BP neural network to generate false samples corresponding to the real samples.

[0071] S303: Input the real samples, preprocessed signal samples, and false samples into the convolutional neural network to determine the authenticity of the samples and classify the samples.

[0072] In one embodiment of the present application, the randomly generated vector Mapped into the BP neural network to generate false samples corresponding to real samples The specific process can be expressed as:

[0073]

[0074] Among them, f is the activation function of tanh, θ z ,θ′ z and θ″ Z are the model parameters from the input layer to the first hidden layer, from the first hidden layer to the second hidden layer, and from the second hidden layer to the output layer. More specifically, W is the weight matrix and bz is the bias vector.

[0075] Compared with BP neural network, CNN needs to complete two tasks during training, namely, judging the authenticity of samples and classifying samples. The features in the Flatten layer are used to identify real samples and fake samples. and They represent the recognition results of real samples, preprocessed samples and fake samples respectively.

[0076] During training, CNN tries to ensure and After adding the data preprocessing process, the corresponding loss function Ld is rewritten. The specific formula is as follows:

[0077]

[0078] in, Indicates that random interference is performed during the training process to improve the robustness of the model and to improve the recognition ability of CNN by maximizing. In contrast, the purpose of BP neural network is to make The corresponding loss function Lg is shown in the following formula.

[0079]

[0080] Here, the deception ability of the BP neural network can be enhanced by maximizing the loss function Lg. The cross entropy function Lc of its category is expressed as follows:

[0081]

[0082] in, and is the category label of the real sample and the false sample through the Softmax layer, and the target labels of the real sample and the false sample are respectively expressed as and

[0083] Considering that ADCNN needs to distinguish the authenticity of samples and classify them, the loss function of the generator G in ADCNN and the loss function of the discriminator D in ADCNN are as follows:

[0084]

[0085] Among them, Θ g and θ d are the parameter sets of BP neural network and CNN respectively.

[0086] In one embodiment of the present application, inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training further includes:

[0087] S401: Fix the parameters of the BP neural network and optimize the loss function of the convolutional neural network discriminator.

[0088] S403: Fix the parameters of the convolutional neural network and optimize the loss function of the convolutional neural network generator.

[0089] In one embodiment of the present application, the goal of CNN is to ensure The goal of the BP neural network is to ensure In order to make the confrontation process between CNN and BP neural network proceed in an orderly manner, the training method of alternating training of CNN and BP neural network is adopted. First, CNN is trained with BP neural network fixed, and L is optimized. ADCNN_D To maximize the diagnostic accuracy of ADCNN. Then, train the B neural network with fixed CNN to optimize L ADCNN_GTo minimize the diagnostic accuracy of ADCNN. The above adversarial process continues until the Nash equilibrium is reached.

[0090] In one embodiment of the present application, the method further includes:

[0091] The t-SNE technique is used to visualize the features of the input layer, the first pooling layer, the second pooling layer, and the Softmax layer of the adversarial denoising convolutional neural network model.

[0092] In one embodiment of the present application, in order to intuitively explain the feature learning process of ADCNN, the features of the input layer, the first pooling layer, the second pooling layer and the Softmax layer of ADCNN are visualized using t-SNE technology. Taking the vibration acceleration signal as input, the experimental results are as follows: Figure 8 As shown, (a) is the original signal, (b) is the first pooling layer, (c) is the second pooling layer, and (d) is the Softmax layer. The distribution of the input data is relatively scattered, and there is aliasing between different categories. After two layers of convolution and pooling, ADCNN gradually mines the features of the input data, and the clustering effect of the data is significantly enhanced. Finally, in the Softmax layer, the inter-class distances of the nine categories are well distinguished, with only partial overlap between category 2 and category 8, and a small number of shared points between other categories. This shows that ADCNN has excellent capabilities in feature mining and feature expression. The changes in adversarial loss and classification loss at each traversal are shown as follows: Fig. 9 (a) and Fig. 9 (b) As shown in Figure 2, it can be seen that as the number of traversals increases, the loss of the model tends to stabilize and eventually converges.

[0093] S207: After preprocessing the vibration signal to be diagnosed, the vibration signal is input into the trained adversarial denoising convolutional neural network model to output the rotating machinery fault diagnosis result.

[0094] In the embodiment of the present application, the trained model is tested using samples with different signal-to-noise ratios (SNR), different sample sizes, and different working conditions, and the performance of the model is evaluated based on the diagnosis results. Noise samples with different signal-to-noise ratios (SNR) are generated by adding Gaussian white noise to the noise-free samples to better simulate the real noise environment. During the experiment, the model is trained using noise-free samples, and the trained model is tested using samples with different SNRs. The definition of SNR is as follows:

[0095]

[0096] Among them, P signal and P noise represent signal power and noise power respectively.

[0097] CNN, GAN-CNN, and some common models such as deep neural network (DNN), stacked autoencoder (SAE), and support vector machine (SVM) were selected as comparison models. To further verify the effectiveness of dynamic erasure, all models were trained using samples processed by the moving maximum method to ensure that there was only a single variable in the entire experimental process. Subsequently, samples with different SNRs were input into the trained model for testing.

[0098] Based on the analysis of the diagnostic results of different models under different signal-to-noise ratios (SNR), the diagnostic results when the SNR is -12dB are shown in Table 3. Thanks to the adversarial mechanism, the diagnostic accuracy of GAN-CNN is improved by 16.46% compared with CNN; and with its denoising performance, the diagnostic accuracy of ADCNN is further improved by 10.34% compared with GAN-CNN. The comparison results show that ADCNN has the highest diagnostic accuracy.

[0099] Table 3. Diagnostic structures of different models when SNR is -12

[0100]

[0101] In the above-mentioned rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network, first, the vibration signal is obtained and preprocessed to obtain the preprocessed signal sample; then, the adversarial denoising convolutional neural network model is constructed based on the BP neural network and the convolutional neural network; then, the preprocessed signal sample and the real sample are input into the adversarial denoising convolutional neural network model for alternating training; finally, after the vibration signal to be diagnosed is preprocessed, it is input into the trained adversarial denoising convolutional neural network model to output the rotating machinery fault diagnosis result. In other words, by introducing the moving maximum method, the anti-noise ability of the training sample is significantly enhanced, and the dynamic erasing technology is combined to simulate noise interference, which helps the model to more accurately capture the fault characteristics in the noise environment, thereby improving the diagnostic accuracy and robustness, and by using the adversarial mechanism to generate more representative training samples, expand the sample size, and enhance the generalization ability of the model, so that it can maintain high diagnostic accuracy under limited sample conditions. By combining adversarial mechanisms, denoising technology and deep learning frameworks, the fault diagnosis performance of rotating machinery is significantly improved in strong noise environments, limited sample conditions and multi-operating condition adaptability, providing a practical and innovative solution for the health monitoring and fault prediction of rotating machinery in the industrial field.

[0102] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0103] Based on the same inventive concept, the embodiment of the present application also provides a rotating machinery fault diagnosis device based on an adversarial denoising convolutional neural network for implementing the above-mentioned rotating machinery fault diagnosis method based on an adversarial denoising convolutional neural network. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of a rotating machinery fault diagnosis device based on an adversarial denoising convolutional neural network provided below can refer to the above limitations on a rotating machinery fault diagnosis method based on an adversarial denoising convolutional neural network, and will not be repeated here.

[0104] In one embodiment, Fig.10 As shown, a rotating machinery fault diagnosis device 1000 based on an adversarial denoising convolutional neural network is provided, comprising: a data preprocessing module 1001, a model building module 1003, a model training module 1005 and a fault diagnosis module 1007, wherein:

[0105] The data preprocessing module 1001 is used to acquire and preprocess the vibration signal to obtain a preprocessed signal sample.

[0106] The model building module 1003 is used to build an adversarial denoising convolutional neural network model based on the BP neural network and the convolutional neural network.

[0107] The model training module 1005 is used to input the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training.

[0108] The fault diagnosis module 1007 is used to pre-process the vibration signal to be diagnosed, input it into the trained adversarial denoising convolutional neural network model, and output the rotating machinery fault diagnosis result.

[0109] In one embodiment of the present application, the data preprocessing module is further used to:

[0110] The vibration signal is processed by using a fast Fourier transform and a moving maximum method to obtain a processed spectrum.

[0111] In one embodiment of the present application, the data preprocessing module is further used to:

[0112] The processed spectrum is dynamically erased using an optimized erasing coefficient.

[0113] In one embodiment of the present application, the model training module is also used to:

[0114] Map the random vector into the BP neural network to generate false samples corresponding to the real samples;

[0115] The real samples, preprocessed signal samples and false samples are input into the convolutional neural network to determine the authenticity of the samples and classify the samples.

[0116] In one embodiment of the present application, the model training module is also used to:

[0117] Fixing the parameters of the BP neural network and optimizing the loss function of the convolutional neural network discriminator;

[0118] The parameters of the convolutional neural network are fixed, and the loss function of the convolutional neural network generator is optimized.

[0119] In one embodiment of the present application, the method further includes:

[0120] The t-SNE technique is used to visualize the features of the input layer, the first pooling layer, the second pooling layer, and the Softmax layer of the adversarial denoising convolutional neural network model.

[0121] Each module in the above-mentioned rotating machinery fault diagnosis device based on adversarial denoising convolutional neural network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0122] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.11As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a rotating machinery fault diagnosis method based on an adversarial noise reduction convolutional neural network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse.

[0123] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0124] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0125] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0126] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0127] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0128] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0129] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network, characterized in that: The method comprises: Acquire the vibration signal and perform preprocessing to obtain a preprocessed signal sample; Construct an adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network; Inputting the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training; After preprocessing, the vibration signal to be diagnosed is input into the trained adversarial denoising convolutional neural network model to output the rotating machinery fault diagnosis result.

2. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network according to claim 1, characterized in that: The pre-processing comprises: The vibration signal is processed by using a fast Fourier transform and a moving maximum method to obtain a processed spectrum.

3. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network according to claim 2, characterized in that: The pre-processing further comprises: The processed spectrum is dynamically erased using an optimized erasing coefficient.

4. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network according to claim 1, characterized in that: The step of inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training comprises: Map the random vector into the BP neural network to generate false samples corresponding to the real samples; The real samples, preprocessed signal samples and false samples are input into the convolutional neural network to determine the authenticity of the samples and classify the samples.

5. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network according to claim 4, characterized in that: The step of inputting the preprocessed signal samples and the real samples into the adversarial denoising convolutional neural network model for alternating training also includes: Fixing the parameters of the BP neural network and optimizing the loss function of the convolutional neural network discriminator; The parameters of the convolutional neural network are fixed, and the loss function of the convolutional neural network generator is optimized.

6. A rotating machinery fault diagnosis method based on adversarial denoising convolutional neural network according to claim 1, characterized in that: The method further comprises: The t-SNE technique is used to visualize the features of the input layer, the first pooling layer, the second pooling layer, and the Softmax layer of the adversarial denoising convolutional neural network model.

7. A rotating machinery fault diagnosis device based on adversarial denoising convolutional neural network, characterized in that: The device comprises: A data preprocessing module is used to obtain and preprocess the vibration signal to obtain a preprocessed signal sample; Model building module, used to build an adversarial denoising convolutional neural network model based on BP neural network and convolutional neural network; A model training module, used for inputting the preprocessed signal samples and real samples into the adversarial denoising convolutional neural network model for alternating training; The fault diagnosis module is used to pre-process the vibration signal to be diagnosed, input it into the trained adversarial denoising convolutional neural network model, and output the rotating machinery fault diagnosis result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images