Rotation machinery fault diagnosis method and device based on audible sound signals

Abstract

The invention discloses a rotating machinery fault diagnosis method and device based on audible sound signals. The method comprises the following steps: preprocessing an obtained original audible sound signal of the rotary mechanical equipment by using a sliding window mode to obtain an original audible sound signal sample set; marking a corresponding label on each sample in the original audible sound signal sample set by adopting a multi-label coding mode; performing Mel spectrogram processing on the original audible sound signal sample set to generate an input data set; performing sample data enhancement on the input data set to obtain a balanced data set; according to a pre-constructed neural network model, training the balanced data set and the corresponding label to obtain a fault classification model, the neural network model being composed of a plurality of convolutional layers and a capsule network layer; and performing fault identification on the to-be-identified audible sound signal data of the to-be-identified rotating mechanical equipment by adopting the fault classification model, and determining a fault classification result of the to-be-identified rotating mechanical equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fault diagnosis of rotating machinery, and more specifically, to a method and device for diagnosing fault of rotating machinery based on audible sound signals. Background Art

[0002] In the context of smart industry, modern rotating machinery is developing towards complexity, precision, efficiency and intelligence. Therefore, research on intelligent fault diagnosis (IFD) has become key to ensure the safe, efficient and reliable operation of such machinery. In recent years, data-driven methods using deep learning have attracted widespread attention for the development of effective intelligent monitoring and fault diagnosis models, however, these methods mainly rely on contact measurements based on vibration analysis. Compared with traditional vibration-based fault diagnosis methods, acoustic-based intelligent fault diagnosis (AIFD) of rotating machinery uses microphones to capture acoustic signals, which has several obvious advantages. First, AIFD can achieve non-contact and non-intrusive measurements without disassembling the equipment. Second, it shows strong adaptability in extreme environments, such as high temperature or corrosive environments. Third, the signal acquisition process of AIFD is not affected by the surface condition of the equipment, and acoustic signals can be captured even on oily, smooth or geometrically irregular surfaces. Finally, AIFD allows continuous diagnosis without interrupting the operation of the machinery, because the installation and diagnosis process does not interfere with normal operation. In addition, as a key technological driver of the fourth industrial revolution, AIFD has become a very promising rotating machinery fault diagnosis technology in recent years. It should be noted that most of the existing literature on AIFD focuses on the occurrence of single faults in rotating machinery. However, in actual industrial environments, the internal structure of rotating machinery is usually complex and the components are tightly coupled. This complexity often leads to multiple faults in a single component or faults distributed across multiple components. AIFD research on such composite faults is relatively limited and needs further exploration.

[0003] Strictly speaking, the challenge of acoustic-based intelligent compound fault diagnosis can be classified as a small sample problem. Compound fault samples are usually difficult and costly to obtain, resulting in datasets dominated by normal operation and single fault categories, with a limited number of compound fault samples. This extreme data imbalance causes the classification model to ignore minority categories, resulting in poor generalization performance for these categories. Data augmentation techniques are considered to be a powerful solution to the problem of limited samples in fault diagnosis. Generative adversarial networks (GANs) provide an effective data augmentation method that uses adversarial training to learn the underlying data distribution from noise to generate more realistic and representative samples, overcoming the limitations of traditional sampling methods. Current research methods highlight the extensive exploration of various GAN architectures in mechanical fault diagnosis, especially addressing the challenges posed by small sample scenarios of vibration signals. However, in the context of AIFD, there is a lack of a comprehensive methodological framework to address the severe challenges posed by the extremely scarce compound fault samples. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a rotating machinery fault diagnosis method and device based on audible sound signals.

[0005] According to one aspect of the present invention, there is provided a rotating machinery fault diagnosis method based on audible sound signals, comprising:

[0006] Preprocessing the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data;

[0007] A multi-label encoding method is used to label each sample in the original audible sound signal sample set with a corresponding label;

[0008] Performing Mel-spectrogram processing on the original audible sound signal sample set to generate an input data set;

[0009] Perform sample data augmentation on the input data set to obtain a balanced data set;

[0010] The balanced data set and its corresponding labels are trained according to the pre-built neural network model to obtain a fault classification model. The neural network model consists of multiple convolutional layers and capsule network layers.

[0011] A fault classification model is used to perform fault identification on the audible sound signal data of the rotating mechanical equipment to be identified, and a fault classification result of the rotating mechanical equipment to be identified is determined.

[0012] Optionally, the original audible sound signal sample set is processed into a mel-spectrogram to generate an input data set, including:

[0013] The original audible sound signal sample set is processed by framing, and a preset window function is added to each frame signal to generate a framed audible sound signal;

[0014] Performing discrete Fourier transform processing on the framed audible sound signal to obtain a discrete signal;

[0015] Calculate the power spectrum of a discrete signal;

[0016] Apply a Mel filter bank to evenly distribute a set of triangular filters on the linear frequency;

[0017] The input data set is generated by multiplying the power spectrum with a triangular filter and taking the logarithm.

[0018] Optionally, the framed audible sound signal f m The calculation formula for (n) is:

[0019] f m (n) = f(n+m·h)·ω(n)

[0020] Where m represents the frame index, n represents the frame size, h represents the frame shift, ω(n) represents the window weight of the frame, and f m (n) represents the frame after adding the window function;

[0021] The calculation formula for discrete signals is:

[0022]

[0023] Where k is the index of the frequency point, K is the number of Fourier transform points, and f m (k) is a discrete Fourier transformed discrete signal corresponding to the frequency component of the frame;

[0024] The power spectrum is calculated as:

[0025] P m (k)=|f m (k)| 2

[0026] The Mel frequency range of the triangle filter is:

[0027]

[0028] Among them, the scale relationship between Mel frequency and sampling frequency is: f is the frequency of the signal, F mel (f) is the Mel frequency, H i (t) represents the response of the Mel filter to frequency, i is the filter index, M is the number of filters, and f(i) is the center frequency of the filter on the Mel frequency scale.

[0029] Optionally, sample data augmentation is performed on the input dataset to obtain a balanced dataset, including:

[0030] Construct an orthogonalized latent space vector containing implicit label information;

[0031] Construct a generative adversarial model that combines latent space control with local perception mechanism. The generative adversarial model includes a generator and a discriminator, where the input of the generator is a latent space vector.

[0032] The generative adversarial model is used to enhance the sample data of the input dataset to obtain a balanced dataset.

[0033] Optionally, the generator includes five deconvolution layers, batch normalization is applied after each deconvolution layer and non-linearity is introduced using the ReLU activation function, and residual blocks are added to the first two layers. The discriminator includes five convolution layers, and the LeakyReLU activation function and batch normalization are introduced, wherein the first layer of the discriminator embeds label information, and the output layer uses a local discriminant perception mechanism for discrimination.

[0034] Optionally, the loss function of the discriminator is:

[0035]

[0036] In the formula, is the interpolation sample between the real data and the generated data, ε is sampled from a uniform distribution, is the gradient of the D output, c is the category label vector, represents the ability of measuring the discriminator to identify real data as real, and measures the ability of the discriminator to identify generated data as fake, λ gp Is the penalty coefficient that controls the weight of the gradient penalty term;

[0037] The loss function of the generator is:

[0038]

[0039] In the formula, φ(·) represents the feature extraction performed by the neural network, x and y are the input real image and the generated image respectively, and λ p is a weight factor used to balance the contribution of adversarial loss and perceptual loss.

[0040] Optionally, the capsule network layer uses a dynamic routing algorithm to determine the coefficient of each capsule transmitting information to the next layer. The coefficient update formula is:

[0041]

[0042] Where b ijrepresents the initial weight between capsule i and capsule j, and C is the total number of capsules in the next layer;

[0043] The nonlinear compression function of the capsule is:

[0044]

[0045] in, is the prediction vector routed from capsule i to capsule j, K l is the total number of capsules i, d j represents the total weighted input from capsule i to capsule j in all previous layers, v j The output of each capsule is trained on the balanced dataset and its corresponding labels according to the pre-built neural network model to obtain the fault classification model, including:

[0046] Optionally, the loss function of the neural network model is:

[0047] L k =T k max(0,m + -||v k ||) 2 +λ·(1-T k )·max(0,||v k ||-m - ) 2

[0048] Where, T k A binary indicator representing a label; ||v k || represents the length of the kth capsule output vector; m + is the boundary value of the positive class; m - is the boundary value of the negative class; λ is the balance coefficient.

[0049] According to another aspect of the present invention, there is provided a rotating machinery fault diagnosis device based on audible sound signals, comprising:

[0050] A preprocessing module, used to preprocess the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data;

[0051] A label encoding module, used for labeling each sample in the original audible sound signal sample set with a corresponding label by using a multi-label encoding method;

[0052] A signal representation module, used for performing mel-spectrogram processing on the original audible sound signal sample set to generate an input data set;

[0053] The data enhancement module is used to enhance the sample data of the input data set to obtain a balanced data set;

[0054] A model training module is used to train the balanced data set and its corresponding labels according to a pre-built neural network model to obtain a fault classification model. The neural network model consists of multiple convolutional layers and capsule network layers.

[0055] The fault identification module is used to use a fault classification model to perform fault identification on the audible sound signal data of the rotating mechanical equipment to be identified, and determine the fault classification result of the rotating mechanical equipment to be identified.

[0056] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above aspects of the present invention.

[0057] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of the above aspects of the present invention.

[0058] Therefore, the present invention proposes a new method for finite sample composite fault diagnosis of rotating machinery based on audible sound signals. First, the original audible sound signal of the rotating machinery is obtained, and a sample set of the original audible sound signal containing a small amount of composite fault data is constructed; then, the Mel spectrum is preprocessed to obtain a training set as the input for training the generative adversarial model; a LP-LSGAN generative adversarial model combining latent space control with local perception mechanism is constructed to control the generation of the composite fault Mel spectrum, enhance the composite fault samples in the training set, and obtain a balanced training set; a CapNet model is constructed, and the balanced data set is input into the CapNet model for training, and multi-label decoupling classification of the fault type is performed to achieve accurate prediction of the fault state. Based on audible sound signals, the present invention proposes a new intelligent fault diagnosis method, which expands the composite fault types in the training set by generating high-resolution Mel spectrum, effectively solving the problem of extremely scarce composite fault samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0060] Figure 1 is a flow chart of a rotating machinery fault diagnosis method based on audible sound signals provided by an exemplary embodiment of the present invention;

[0061] Figure 2is another flow chart of a rotating machinery fault diagnosis method based on audible sound signals provided by an exemplary embodiment of the present invention;

[0062] Figure 3 is a Mel spectrum acquisition flow chart provided by an exemplary embodiment of the present invention;

[0063] Figure 4 is a detailed structural diagram of a LP-LSGAN generative adversarial model provided by an exemplary embodiment of the present invention;

[0064] Figure 5 is a fault diagnosis flow chart of multi-label classification of the CapNet model provided by an exemplary embodiment of the present invention;

[0065] Figure 6 is a structural schematic diagram of a rotating machinery fault diagnosis device based on audible sound signals provided by an exemplary embodiment of the present invention;

[0066] Figure 7 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0067] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0068] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0069] Those skilled in the art can understand that the terms "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor indicate the necessary logical order between them.

[0070] It should also be understood that, in the embodiments of the present invention, “plurality” may refer to two or more than two, and “at least one” may refer to one, two or more than two.

[0071] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0072] In addition, the term "and / or" in the present invention is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects before and after are in an "or" relationship.

[0073] It should also be understood that the description of the various embodiments of the present invention focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced to each other, and for the sake of brevity, they will not be described one by one.

[0074] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0075] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0076] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0077] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0078] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, etc.

[0079] Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system executable instructions (such as program modules) executed by computer systems. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in a distributed cloud computing environment, where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0080] Exemplary Methods

[0081] Figure 1 FIG. 1 is a flow chart of a rotating machinery fault diagnosis method based on audible sound signals provided by an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as Figure 1 As shown, the rotating machinery fault diagnosis method 100 based on audible sound signals includes the following steps:

[0082] Step 101, preprocessing the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data;

[0083] Step 102, using a multi-label encoding method to label each sample in the original audible sound signal sample set with a corresponding label;

[0084] Step 103, performing Mel-spectrogram processing on the original audible sound signal sample set to generate an input data set;

[0085] Step 104, performing sample data enhancement on the input data set to obtain a balanced data set;

[0086] Step 105, training the balanced data set and its corresponding labels according to the pre-built neural network model to obtain a fault classification model, where the neural network model is composed of multiple convolutional layers and capsule network layers;

[0087] Step 106 , using a fault classification model to perform fault identification on the audible sound signal data of the rotating mechanical device to be identified, and determining a fault classification result of the rotating mechanical device to be identified.

[0088] Specifically, the present invention discloses a new method for diagnosing finite sample composite faults of rotating machinery based on audible sound signals. It includes a Mel spectrum graph for feature representation, a latent space controlled generative adversarial network (LP-LSGAN) with a local perception mechanism for generating high-quality composite fault samples and a multi-label classification model based on CapNet. The present invention provides an effective intelligent fault diagnosis solution by solving the problem of accuracy deviation caused by the imbalance of training samples caused by the scarcity of composite faults.

[0089] The technical solution adopted by the present invention is:

[0090] Step 1): Use envelope surface measurement to obtain audible sound signals of multiple preset components of rotating mechanical equipment under preset working conditions, organize audible sound signal data sets of single faults and compound faults, and label each obtained sample with corresponding multiple labels.

[0091] Step 2): Perform non-overlapping sampling on the above signals and use the Mel spectrum to convert the signals into two-dimensional grayscale images to enhance the fault feature representation of the audible sound signal. Then, normalize these images, and use a large number of single faults and a small number of compound faults as training samples, and a sufficient number of single faults and compound faults as test samples.

[0092] Step 3): Construct an orthogonalized latent space vector containing implicit label information; and develop an LP-LSGAN generative adversarial model that combines latent space control with local perception mechanism. Input the training samples obtained in step 2) into LP-LSGAN for adversarial training, and use LP-LSGAN to accurately control the output of the composite fault Mel spectrum map.

[0093] Step 4): Use the LP-LSGAN generative adversarial model trained in step 3) to generate a large number of compound fault samples to expand the limited training sample set. Use the generated samples to assist in training the CapNet deep learning model for accurate classification and decoupling of compound fault types.

[0094] It should be further explained that in step 1), a multi-label classification method is used to encode the fault type. The multi-label encoding method is similar to binary encoding, where each bit represents the presence (1) or absence (0) of a specific single fault. For example, if fault1 and fault2 exist, their labels are [1,1,0]; if fault1 and fault3 exist, their labels are [1,0,1]. This encoding method effectively captures the relationship between single faults and compound faults through encoding.

[0095] It should be further explained that in step 2), the conversion process of the Mel spectrogram uses the Mel scale to convert the audible sound signal from the time domain to a time-frequency domain form that is more suitable for analysis. It shows obvious sensitivity and fast response in the low-frequency spectrum, which is very close to the ability of human hearing. On the contrary, its sensitivity decreases at higher frequencies, which is also consistent with human auditory perception. In view of these advantages, the Mel scale can analyze and process acoustic signals more accurately and effectively. The process of converting the original time domain signal to the Mel spectrogram using the Mel scale is as follows:

[0096] First, the audible sound signal f(c) is divided into overlapping frames, where c is the sampling point and c = 1, 2, ..., N. Then, frame shifting is used to control the degree of overlap between adjacent frames. A window function is added to each frame to obtain the framed audible sound signal f m (n):

[0097] f m (n) = f(n+m·h)·ω(n)

[0098] Where m represents the frame index, n represents the frame size, h represents the frame shift, ω(n) represents the window weight of the frame, and f m (n) represents the frame after adding the window function.

[0099] Next, the windowed frame is subjected to discrete Fourier transform (DFT):

[0100]

[0101] Where k is the index of the frequency point, K is the number of Fourier transform points, and f m (k) is the result of DFT corresponding to the frequency components of the frame.

[0102] Then calculate the power spectrum P according to the DFT result m (k)=|f m (k)| 2 .

[0103] The scale relationship between Mel frequency and sampling frequency is as follows:

[0104]

[0105] Where f is the frequency of the signal, F mel (f) is the Mel frequency.

[0106] According to the Mel frequency scale, the Mel filter bank is designed to correspond to a specific Mel frequency range. The calculation method is as follows:

[0107]

[0108] In the formula, H i (k) represents the response of the Mel filter to frequency, i is the filter index, M is the number of filters, and f(i) is the center frequency of the filter on the Mel frequency scale.

[0109] The Mel spectrum is obtained by multiplying the power spectrum with the Mel filter bank and then taking the logarithm:

[0110]

[0111] Where t represents the time frame and f represents the Mel frequency.

[0112] It should be further explained that in the step 3), it is first necessary to construct an orthogonalized latent space vector containing implicit label information. In order to represent different composite fault attributes, a d-dimensional orthogonalized latent space vector is designed, which obeys the Gaussian distribution. Assuming that there are three single fault types, the vector is divided into four sections: the first part represents the normal operation of the rotating machinery, with an interval of 0 to d / 4; the second part represents the occurrence of a single fault type fault1, with an interval of d / 4 to d / 2; similarly, d / 2 to 3d / 4 represents the occurrence of fault2; 3d / 4 to d represents the occurrence of fault3. The division of the vector attribute expression interval can make the latent space expression vectors of the three single faults completely orthogonal, relatively independent, and unaffected by each other. It is even possible to generate unprecedented bearing composite fault categories through numerical operations on these intervals.

[0113] It should be further explained that in the step 3), a LP-LSGAN generative adversarial model combining latent space control with local perception mechanism is developed, and the model mainly includes two parts: generator and discriminator. The generator is designed with five deconvolution layers, and batch normalization (BatchNorm) is applied after each deconvolution layer to stabilize the training process, and nonlinearity is introduced using the ReLU activation function. In addition, residual blocks are introduced in the first two deconvolution layers. These blocks help to retain and transmit label information to deeper layers of the generator, thereby reducing the risk of information loss or distortion during intermediate convolution operations. And the discriminator introduces the discriminative idea of ​​the local perception mechanism, and judges the local area (Patch) of the image block by block to generate a higher resolution image. The discriminator has 5 convolution layers and uses the LeakyReLU activation function to alleviate the possible gradient vanishing problem. At the same time, BatchNorm is applied to stabilize the training process and accelerate convergence. In addition, label information is embedded in the first layer of the discriminator to combine the category information to make the discriminator make more accurate judgments.

[0114] It should be further explained that in step 3), in the process of training the above LP-LSGAN generative adversarial model, the Wasserstein distance is used as the loss function to measure the difference between the generated data and the real data for iteration, and the gradient penalty term (GP) is introduced to prevent gradient explosion. The loss formula of the final discriminator is calculated as follows:

[0115]

[0116] In the formula, is the interpolation sample between the real data and the generated data, ε is sampled from a uniform distribution, yes D The gradient of the output. Where c is the class label vector. represents the ability of measuring the discriminator to identify real data as real, and Measures the ability of the discriminator to identify generated data as fake. gp is the penalty coefficient that controls the weight of the gradient penalty term.

[0117] In addition to adversarial loss, perceptual loss can further improve the quality of generated images. Perceptual loss is based on pre-trained CNNs, such as VGG, which captures higher-level visual features by measuring the difference between generated images and real images in feature space. Therefore, the loss formula of the final generator is calculated as follows:

[0118]

[0119] Where φ(·) represents the feature extraction performed by the VGG network, x and y are the input real image and the generated image, respectively, and λ p is a weight factor used to balance the contribution of adversarial loss and perceptual loss.

[0120] It should be further explained that in step 4), the samples generated by LP-LSGAN are used to assist in training the CapNet deep learning model for accurate classification and decoupling of compound fault types. The classification model uses a dynamic routing algorithm to determine how much information each capsule should pass to the next layer. Initially, the coupling coefficients (indicating the connection strength between one layer of capsules and the next layer of capsules) are randomly initialized. These coefficients are iteratively updated using the softmax function:

[0121]

[0122] Where b ij It represents the initial weight between capsule i and capsule j, and the total number of capsules in the next layer is C.

[0123] The routing algorithm iteratively adjusts these coupling coefficients based on the consistency between the prediction of the current capsule and the output of the next layer. This consistency is expressed by the vector v j and It is measured by the scalar product between , where is the output vector of capsule j and is The prediction vector from capsule route i to capsule j.

[0124]

[0125] Where, d j Represents the total weighted input of capsule j in the current layer.

[0126] To ensure that the capsule output is within a reasonable range, a nonlinear compression function is applied to the capsule output vector.

[0127]

[0128] This function compresses short vectors to a length close to zero and ensures that longer vectors are close to 1, helping the network focus on the most relevant features. The final output of the decoupled classifier is converted into a prediction vector y pred , where each element y i Represents the probability that the input belongs to category i. The classifier uses the L2 norm to refine the output tensor and uses the threshold to determine the predicted category. If y i If the threshold γ is exceeded, the sample is classified as belonging to that class; otherwise, it does not belong to that class. This process ensures reliable predictions by distinguishing features that are clearly associated with each class.

[0129] It should be further explained that in step 4), the CapNet classification method uses the marginal loss function as the loss function for iteration. The marginal loss function can be used in capsule networks to effectively handle multi-label problems. For each label, the formula of the marginal loss function can be expressed as:

[0130] L k =T k max(0,m + -||v k ||) 2 +λ·(1-T k )·max(0,||v k ||-m - ) 2

[0131] Where, T k represents a binary indicator of the label. If the true class is k, then T k =1, otherwise T k =0. ||v k|| represents the length (modulus) of the kth capsule output vector. m + It is the boundary value of the positive class, indicating that the correctly classified samples should have a larger output. - is the boundary value of the negative class, indicating that the misclassified samples should have a smaller output. λ is a balancing coefficient used to control the weight of the negative class loss, which usually takes a smaller value.

[0132] In one embodiment of the present invention, the specific implementation process is as follows:

[0133] Step 1): Use the envelope surface measurement method to obtain the audible sound signals of multiple preset components of the rotating mechanical equipment under preset working conditions, and use the sliding window to preprocess the data. Apply non-overlapping segmentation to the signal to obtain the original audible sound signal sample set, including a large amount of single fault data and a small amount of compound fault data. In order to illustrate the implementation method of the embodiment of the present application, this embodiment uses a planetary gearbox as a rotating mechanical experimental object, and processes three different single fault types (cracks, pitting, broken teeth) and four compound fault types on the sun gear. And the torque of the magnetic powder brake is set to 1N·m, and the speed of the variable frequency motor is set to 1800rpm. Then use the acoustic test system to sample at a sampling frequency of 10kHz to obtain the original audible sound signal. By setting the length of the sliding window to 4096 for non-overlapping segmentation, a sample set of 3000 samples of each fault type is obtained.

[0134] Step 2): The original audible sound signal sample set obtained in step 1) is processed into a Mel spectrum diagram, such as Figure 2 As shown. First, load the faulty audio data from the file, then divide the frames and use the window function to add an audio window operation to each frame. Then perform a discrete Fourier transform on each frame signal to calculate its graph, and then square the spectrum of each frame to obtain the power spectrum. Then apply the Mel filter group to evenly distribute a group of triangular filters on the linear frequency. Finally, in order to simulate the human ear's perception of sound, the energy of the Mel filter group is logarithmically transformed to finally obtain the Mel spectrum graph. Taking the above embodiment as an example, the frame size is set to 1024, the frame shift is set to 64, and the window function uses a Hanning window. After processing according to the Mel spectrum graph processing process, a 64×64 Mel grayscale image will be obtained, which is used as the final input data set for generating adversarial model training.

[0135] In step 2), the 3000 Mel image samples of each category are further divided into a training set and a test set. In order to evaluate the effectiveness of fault diagnosis after data enhancement when compound fault samples are very limited, only 5 Mel images are used for each compound fault type to join the training set. 1500 Mel images are used for each single fault type and normal operation class to join the training set. The test set includes 1500 samples of each type of normal operation, single fault type and compound fault type.

[0136] Step 3): Use an LP-LSGAN generation model that combines latent space control with local perception mechanism to control the generation of compound fault Mel spectrum graphs and enhance the compound fault samples in the training set. The generator and discriminator are the two core modules of LP-LSGAN, such as Figure 3 As shown, the structures of these two core modules are described step by step according to this embodiment.

[0137] In the step 3), the input of the generator module is a 256-dimensional orthogonalized random noise vector, which represents the latent space vector. The design of the latent space vector includes first initializing a 256-dimensional random noise vector that obeys a Gaussian distribution, and then dividing the vector into four parts: the first part represents the normal operation of the rotating machinery, with an interval of 0 to 63; the second part represents the occurrence of a single fault type fault1, with an interval of 64 to 127; similarly, 128 to 191 represents the occurrence of fault2; 192 to 255 represents the occurrence of fault3. If any type of fault exists, the value of this part is maintained; if not, the value of this part is set to zero. In this way, the corresponding latent space vector can be obtained according to the fault type label in the training set. For example, the [1,1,0] multi-label type represents a composite fault type in which fault1 and fault2 exist at the same time, and the corresponding latent space vector is a random noise vector with 0 to 191 bits and 0 with 192 to 256 bits.

[0138] In the step 3), the latent space vector after the orthogonalization process is used as the input of the generator. Each layer of the generator consists of a convolutional transpose layer (ConvTranspose2d), batch normalization (BatchNorm) and an activation function (ReLU). The design gradually upsamples the latent space vector and converts it into the target Mel grayscale image obtained in step 2). In addition, a residual block (ResidualBlock) is added to the first two layers. The residual block contains two Conv2d operations with a kernel size of 3×3 and a stride of 1×1, and each convolution operation is followed by a BatchNorm and ReLU activation function. The residual block retains the input information through jump connections and performs deep feature extraction. The specific details of the generator are shown in the following table:

[0139]

[0140] In the step 3), the input image and label are concatenated and used as the input of the discriminator. The structure of the discriminator consists of multiple convolutional layers (Conv2d), and BatchNorm and LeakyReLU activation functions are added after most of the convolutional layers to improve the stability and nonlinear expression ability of the network. This design extracts local features of images and labels layer by layer, and finally outputs a result indicating true or false discrimination. In addition, the discriminator adopts a local discrimination (Patch) method, and obtains the discrimination results of the local area of ​​the image through a convolutional layer of size 1×1 in the output layer, further improving the discrimination ability of the model. The specific details of the discriminator are shown in the following table:

[0141]

[0142]

[0143] In step 3), Wasserstein distance is used as the loss function to measure the difference between generated data and real data, and a gradient penalty term (GP) is introduced to prevent gradient explosion. The loss function of the generator consists of adversarial loss and perceptual loss. Perceptual loss calculates the difference between images in a higher-level feature space and is implemented through a pre-trained model (VGG16). p Set to 0.5. Finally, the loss of the discriminator is gp Set to 10.

[0144] In step 3), the training set obtained in step 2) is used to input the above-constructed LP-LSGAN generative adversarial model for training. The Adam optimizer is used, and the batch size of each experiment is set to 512; the learning rate of the generative model and is set to 0.0001, and multiple epochs are trained until the Nash equilibrium is reached. Finally, the trained LP-LSGAN is used to accurately control the output of the composite fault Mel spectrum map, and is added to the original training set as an enhanced data set to balance the training data.

[0145] Step 4): Use the balanced training dataset obtained in step 3) as the input of the CapNet deep learning model, such as Figure 4 The figure shows the architecture details of the CapNet deep learning model. The model is trained and multi-label classification and decoupling of fault types are performed in the test set to obtain the final fault diagnosis results. In this embodiment, the learning rate of the CapNet model is set to 0.0001 during the training process, and the learning rate is adjusted using the step decay method, and the training is performed for 100 epochs.

[0146] In step 4), the CapNet model consists of multiple convolutional layers and capsule network layers, combining the extraction of spatial information with the feature processing capabilities of the capsule layer, and further optimizes the classification results through a dynamic routing mechanism. First, the input image passes through a convolutional layer (Conv2d), which receives a single-channel input and outputs a feature map of 256 channels, with a convolution kernel size of 5×5, a stride of 2, and a padding of 1. Nonlinear transformations are introduced through the ReLU activation function. Next is the PrimaryCaps layer, which outputs a feature map of 8×32 (i.e., 256 capsules, 32 channels, and each capsule contains 8 features) through a convolution operation (Conv2d), with a convolution kernel size of 9×9 and a stride of 2. This layer converts the convolution feature map into a capsule representation, allowing more complex patterns and spatial information to be captured. The DigitCaps layer consists of 3 linear layers, and the linear transformation in each capsule layer is responsible for generating a 16-dimensional capsule feature vector. In this way, the model can capture different category features of the input image. The dynamic routing mechanism is iteratively executed between capsules to ensure that high-confidence outputs are passed to the target capsule layer. In this embodiment, since it is assumed that there are three single fault types in the embodiment, the final capsule output vector is flattened to obtain three independent feature vectors. Each feature vector represents a classification label, so this model is designed as a multi-label classification problem, in which the model predicts the category probabilities of three labels at the same time. The threshold is set to 0.5 here, that is, when the predicted probability value is greater than 0.5, the label position is 1; when it is less than 0.5, the label position is 0.

[0147] In step 4), the CapNet model constructed above is used to train the balanced training data set, and the marginal loss function is used during the training process to measure the difference between the model output features and the true labels. The marginal loss function can effectively handle multi-label problems when used in capsule networks. In the embodiment, the boundary value of the positive class in the marginal loss function is set to 0.9, the boundary value of the negative class is set to 0.1, and the weight for controlling the negative class loss is set to 0.25.

[0148] In summary, the framework of the new method for finite sample composite fault diagnosis of rotating machinery based on audible sound signals of the present invention mainly consists of three parts: Mel spectrum processing, LP-LSGAN generative adversarial network to generate Mel spectrum corresponding to composite faults and CapNet model for multi-label decoupling classification. By starting from the three aspects of audible sound signal characterization, balanced data set and multi-label decoupling, the corresponding processing method is designed to improve the diagnostic accuracy of composite fault diagnosis under finite samples based on audible sound signals.

[0149] Compared with the prior art, the present invention has the following beneficial effects:

[0150] The present invention introduces speech signal processing technology and adopts Mel spectrum mapping containing rich time-frequency information as a quantitative representation of audible sound signals to effectively capture and characterize the characteristics of different fault types.

[0151] Aiming at the data imbalance problem caused by limited composite fault samples, this paper proposes a latent space controlled LP-LSGAN generative adversarial model with local perception mechanism. This model can generate high-quality image data that retains local details and global features, effectively alleviating the limitations of unbalanced datasets.

[0152] In order to decouple composite faults, the present invention introduces the CapNet multi-label decoupling classifier. The classifier outputs multiple labels for samples representing composite faults, thereby overcoming the fact that existing composite fault research usually treats single faults and composite faults as independent fault modes and ignores the potential relationship between them, and achieves more detailed diagnosis.

[0153] Exemplary Devices

[0154] Figure 6 FIG. 1 is a schematic diagram of a rotating machinery fault diagnosis device based on audible sound signals provided by an exemplary embodiment of the present invention. Figure 6 As shown, the device 600 includes:

[0155] A preprocessing module 610 is used to preprocess the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data;

[0156] The label encoding module 620 is used to label each sample in the original audible sound signal sample set with a corresponding label by using a multi-label encoding method;

[0157] The signal characterization module 630 is used to perform mel-spectrogram processing on the original audible sound signal sample set to generate an input data set;

[0158] A data enhancement module 640 is used to perform sample data enhancement on the input data set to obtain a balanced data set;

[0159] A model training module 650 is used to train the balanced data set and its corresponding labels according to a pre-built neural network model to obtain a fault classification model, where the neural network model is composed of multiple convolutional layers and capsule network layers;

[0160] The fault identification module 660 is used to perform fault identification on the audible sound signal data of the rotating mechanical device to be identified by using a fault classification model, and determine a fault classification result of the rotating mechanical device to be identified.

[0161] Optionally, the signal characterization module 630 includes:

[0162] A framing submodule, for framing the original audible sound signal sample set, and adding a preset window function to each frame signal to generate a framed audible sound signal;

[0163] A transform submodule, used for performing discrete Fourier transform processing on the framed audible sound signal to obtain a discrete signal;

[0164] A calculation submodule, used to calculate the power spectrum of the discrete signal;

[0165] A distribution submodule for applying a Mel filter bank to evenly distribute a set of triangular filters on a linear frequency;

[0166] A generation submodule that multiplies the power spectrum with a triangular filter and takes the logarithm to generate the input data set.

[0167] Optionally, the calculation formula of the framed audible sound signal fm(n) is:

[0168] f m (n) = f(n+m·h)·ω(n)

[0169] Where m represents the frame index, n represents the frame size, h represents the frame shift, ω(n) represents the window weight of the frame, and f m (n) represents the frame after adding the window function;

[0170] The calculation formula for discrete signals is:

[0171]

[0172] Where k is the index of the frequency point, K is the number of Fourier transform points, and f m (k) is a discrete Fourier transformed discrete signal corresponding to the frequency component of the frame;

[0173] The power spectrum is calculated as:

[0174] P m (k)=|f m (k)| 2

[0175] The Mel frequency range of the triangle filter is:

[0176]

[0177] Among them, the scale relationship between Mel frequency and sampling frequency is: f is the frequency of the signal, F mel (f) is the Mel frequency, H i(k) represents the response of the Mel filter to frequency, i is the filter index, M is the number of filters, and f(i) is the center frequency of the filter on the Mel frequency scale.

[0178] Optionally, the data enhancement module 640 includes:

[0179] The first construction submodule is used to construct an orthogonalized latent space vector containing implicit label information;

[0180] The second construction submodule is used to construct a generative adversarial model that combines latent space control with a local perception mechanism. The generative adversarial model includes a generator and a discriminator, where the input of the generator is a latent space vector;

[0181] The data enhancement submodule is used to enhance the sample data of the input data set by using the generative adversarial model to obtain a balanced data set.

[0182] Optionally, the generator includes five deconvolution layers, batch normalization is applied after each deconvolution layer and non-linearity is introduced using the ReLU activation function, and residual blocks are added to the first two layers. The discriminator includes five convolution layers, and the LeakyReLU activation function and batch normalization are introduced, wherein the first layer of the discriminator embeds label information, and the output layer uses a local discriminant perception mechanism for discrimination.

[0183] Optionally, the loss function of the discriminator is:

[0184]

[0185] In the formula, is the interpolation sample between the real data and the generated data, ε is sampled from a uniform distribution, is the gradient of the D output, c is the category label vector, represents the ability of measuring the discriminator to identify real data as real, and measures the ability of the discriminator to identify generated data as fake, λ gp Is the penalty coefficient that controls the weight of the gradient penalty term;

[0186] The loss function of the generator is:

[0187]

[0188] In the formula, φ(·) represents the feature extraction performed by the neural network, x and y are the input real image and the generated image respectively, and λ p is a weight factor used to balance the contribution of adversarial loss and perceptual loss.

[0189] Optionally, the capsule network layer uses a dynamic routing algorithm to determine the coefficient of each capsule transmitting information to the next layer. The coefficient update formula is:

[0190]

[0191] Where b ij represents the initial weight between capsule i and capsule j, and C is the total number of capsules in the next layer;

[0192] The nonlinear compression function of the capsule is:

[0193]

[0194] in, is the prediction vector routed from capsule i to capsule j, K l is the total number of capsules i, d j represents the total weighted input from capsule i to capsule j in all previous layers, v j The output of each capsule is trained on the balanced dataset and its corresponding labels according to the pre-built neural network model to obtain the fault classification model, including:

[0195] Optionally, the loss function of the neural network model is:

[0196] L k =T k max(0,m + -||v k ||) 2 +λ·(1-T k )·max(0,||v k ||-m - ) 2

[0197] Where, T k A binary indicator representing a label; ||v k || represents the length of the kth capsule output vector; m + is the boundary value of the positive class; m - is the boundary value of the negative class; λ is the balance coefficient.

[0198] Exemplary Electronic Devices

[0199] Figure 7 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. Figure 7 As shown, the electronic device 70 includes one or more processors 71 and a memory 72 .

[0200] The processor 71 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0201] The memory 72 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 71 may run the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may also include: an input device 73 and an output device 74, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0202] In addition, the input device 73 may also include, for example, a keyboard, a mouse, and the like.

[0203] The output device 74 can output various information to the outside, and can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto.

[0204] Of course, to simplify, Figure 7 Only some of the components related to the present invention in the electronic device are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application conditions, the electronic device may also include any other appropriate components.

[0205] Exemplary computer program products and computer-readable storage media

[0206] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above-mentioned "Exemplary Method" section of this specification.

[0207] The computer program product may be written in any combination of one or more programming languages ​​to write program code for performing the operations of the embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0208] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0209] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0210] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details disclosed above are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0211] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0212] The block diagrams of the devices, systems, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, systems, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0213] The method and system of the present invention may be implemented in many ways. For example, the method and system of the present invention may be implemented by software, hardware, firmware or any combination of software, hardware, firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present invention are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0214] It should also be noted that in the system, device and method of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. The above description of the disclosed aspects is provided to enable any technician in the field to make or use the present invention. Various modifications to these aspects are very obvious to those skilled in the art, and the general principles defined here can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown here, but in accordance with the widest range consistent with the principles and novel features disclosed here.

[0215] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A rotating machinery fault diagnosis method based on audible sound signals, characterized in that: include: Preprocessing the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data; Using a multi-label encoding method to label each sample in the original audible sound signal sample set with a corresponding label; Performing Mel-spectrogram processing on the original audible sound signal sample set to generate an input data set; Performing sample data enhancement on the input data set to obtain a balanced data set; The balanced data set and its corresponding labels are trained according to a pre-built neural network model to obtain a fault classification model, wherein the neural network model is composed of multiple convolutional layers and capsule network layers; The fault classification model is used to perform fault identification on the audible sound signal data of the rotating mechanical equipment to be identified, and a fault classification result of the rotating mechanical equipment to be identified is determined.

2. The method according to claim 1, characterized in that The original audible sound signal sample set is processed into a Mel-spectrogram to generate an input data set, including: Performing frame processing on the original audible sound signal sample set, and adding a preset window function to each frame signal to generate a framed audible sound signal; Performing discrete Fourier transform processing on the framed audible sound signal to obtain a discrete signal; Calculating a power spectrum of the discrete signal; Apply a Mel filter bank to evenly distribute a set of triangular filters on the linear frequency; The power spectrum is multiplied by the triangular filter and the logarithm is taken to generate the input data set.

3. The method according to claim 1, characterized in that The framed audible sound signal f m The calculation formula for (n) is: f m (n)=f(n+m·h)·ω(n) Where m represents the frame index, n represents the frame size, h represents the frame shift, ω(n) represents the window weight of the frame, and f m (n) represents the frame after adding the window function; The calculation formula of the discrete signal is: Where k is the index of the frequency point, K is the number of Fourier transform points, and f m (k) is a discrete Fourier transformed discrete signal corresponding to the frequency component of the frame; The calculation formula of the power spectrum is: P m (k)=|f m (k)| 2 The Mel frequency range of the triangular filter is: Among them, the scale relationship between Mel frequency and sampling frequency is: f is the frequency of the signal, F mel (f) is the Mel frequency, H i (k) represents the response of the Mel filter to frequency, i is the filter index, M is the number of filters, and f(i) is the center frequency of the filter on the Mel frequency scale.

4. The method according to claim 1, characterized in that: Performing sample data augmentation on the input data set to obtain a balanced data set, including: Construct an orthogonalized latent space vector containing implicit label information; Constructing a generative adversarial model that combines latent space control with a local perception mechanism, wherein the generative adversarial model includes a generator and a discriminator, wherein the input of the generator is the latent space vector; The generative adversarial model is used to perform sample data enhancement on the input data set to obtain the balanced data set.

5. The method according to claim 4, characterized in that The generator includes five deconvolution layers, batch normalization is applied after each deconvolution layer and nonlinearity is introduced using the ReLU activation function, and residual blocks are added to the first two layers. The discriminator includes five convolution layers, and the LeakyReLU activation function and batch normalization are introduced, wherein the first layer of the discriminator embeds label information, and the output layer uses a local discriminant perception mechanism for discrimination.

6. The method according to claim 4, characterized in that The loss function of the discriminator is: In the formula, is the interpolation sample between the real data and the generated data, ε is sampled from a uniform distribution, is the gradient of the D output, c is the category label vector, represents the ability of measuring the discriminator to identify real data as real, and measures the ability of the discriminator to identify generated data as fake, λ gp Is the penalty coefficient that controls the weight of the gradient penalty term; The loss function of the generator is: In the formula, φ(·) represents the feature extraction performed by the neural network, x and y are the input real image and the generated image respectively, and λ p is a weight factor used to balance the contribution of adversarial loss and perceptual loss.

7. The method according to claim 1, characterized in that The capsule network layer uses a dynamic routing algorithm to determine the coefficient of each capsule transmitting information to the next layer. The coefficient update formula is: Where b ij represents the initial weight between capsule i and capsule j, and C is the total number of capsules in the next layer; The nonlinear compression function of the capsule is: in, is the prediction vector routed from capsule i to capsule j, K l is the total number of capsules, d j represents the total weighted input from capsule i to capsule j in all previous layers, v j is the output of each capsule.

8. The method according to claim 1, characterized in that The loss function of the neural network model is: L k =T k ·max(0,m + -||v k ||) 2 +λ·(1-T k )·max(0,||v k ||-m - ) 2 Where, T k A binary indicator representing a label; ||v k || represents the length of the kth capsule output vector; m + is the boundary value of the positive class; m - is the boundary value of the negative class; λ is the balance coefficient.

9. A rotating machinery fault diagnosis device based on audible sound signals, characterized in that: include: A preprocessing module, configured to preprocess the acquired original audible sound signal of the rotating mechanical equipment by using a sliding window method to obtain an original audible sound signal sample set, wherein the original audible sound signal sample set includes single fault data and compound fault data; A label encoding module, used for labeling each sample in the original audible sound signal sample set with a corresponding label by using a multi-label encoding method; A signal characterization module, configured to perform Mel-spectrogram processing on the original audible sound signal sample set to generate an input data set; A data enhancement module, used to perform sample data enhancement on the input data set to obtain a balanced data set; A model training module, used to train the balanced data set and its corresponding labels according to a pre-built neural network model to obtain a fault classification model, wherein the neural network model is composed of multiple convolutional layers and capsule network layers; The fault identification module is used to use the fault classification model to perform fault identification on the audible sound signal data of the rotating mechanical equipment to be identified, and determine the fault classification result of the rotating mechanical equipment to be identified.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1 to 8.

Images