A centrifugal pump acoustic radiation signal fault diagnosis method, system and equipment

Abstract

The present invention relates to the technical field of centrifugal pump fault diagnosis and classification, and specifically to a method for fault diagnosis of centrifugal pump acoustic radiation signals, comprising: obtaining a first acoustic radiation signal under different working conditions; filtering the first acoustic radiation signal with a non-recursive filter to obtain a second acoustic radiation signal; converting the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set using a Gram angle field; improving a densely connected convolutional network structure and optimizing the hyperparameter combination of the network structure using a Bayesian optimization algorithm to obtain an acoustic radiation signal fault diagnosis model; and using the acoustic radiation signal fault diagnosis model to perform centrifugal pump fault diagnosis. The present invention converts a one-dimensional time series signal into a two-dimensional image, which can retain the time information of the fault feature and help improve the fault recognition rate. At the same time, the Bayesian optimization algorithm obtains the hyperparameters of the improved densely connected convolutional network, so that the improved densely connected convolutional network has a faster convergence speed and a higher fault recognition accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of centrifugal pump fault diagnosis classification, and in particular relates to a centrifugal pump acoustic radiation signal fault diagnosis method, system and equipment. Background Art

[0002] Centrifugal pumps are widely used in fields such as petrochemicals, energy and power, mining, and national defense, and are crucial to the normal operation of the entire system. Due to their complex operating environments and long, uninterrupted operation, centrifugal pumps often experience various faults, such as cavitation, shaft misalignment, looseness, and bearing damage. These faults not only increase vibration and noise, but can also lead to production stoppages and accidents, resulting in economic losses and safety hazards. Therefore, centrifugal pump fault diagnosis is an issue that urgently needs attention.

[0003] Centrifugal pump fault diagnosis based on vibration signals: Ke Yao et al. performed parallel factor analysis on vibration signals, using time loading factors and frequency loading factors as feature parameters. They then used a support vector machine optimized with an improved particle swarm optimization algorithm to classify normal and faulty states of centrifugal pumps. Fan Chuanhan et al. extracted time-frequency domain features from the vibration signals at the centrifugal pump inlet flange and used random forests to select important features as input to a particle swarm optimization support vector machine, enabling the diagnosis of rotor misalignment and rotor imbalance faults. Zhou Haijun et al. performed wavelet denoising and signal reconstruction on centrifugal pump vibration signals. They used local feature scale decomposition to extract the intrinsic modal components of the reconstructed signals and used their sample entropy as features to train a random forest to diagnose rolling bearing faults in centrifugal pumps. Jiao Hanhui et al. used the time, frequency, energy, and entropy domains of vibration signals as feature datasets, and used a compensated distance evaluation method to achieve feature dimensionality reduction. They then trained a one-dimensional convolutional neural network diagnostic model, achieving high classification accuracy. Liang Xing et al. extracted multifractal spectrum feature parameters from centrifugal pump vibration signals and used these as features to diagnose cavitation faults in centrifugal pumps using a BP neural network. These centrifugal pump fault diagnosis methods are all based on machine learning diagnostic methods that process vibration signal data, extract features, and train models. The frequency range of the acquired signals is much lower than that of acoustic signals. These centrifugal pump fault diagnosis methods do not effectively preserve the temporal information of the fault characteristics.

[0004] Deep learning fault diagnosis based on vibration signals and the progress made by convolutional neural networks (CNNs) in image recognition have driven the development of equipment fault diagnosis. Li et al. constructed convolution kernels with adaptive sizes that match the data source channels and proposed a deep learning-based adaptive data fusion strategy and convolutional neural network-based centrifugal pump fault diagnosis method. Hasan et al. performed a continuous wavelet transform on the centrifugal pump vibration signal to obtain a two-dimensional time-frequency map. This map was further converted into a grayscale image and an adaptive deep convolutional neural network (ADCNN) was used to achieve adaptive feature extraction and identify centrifugal pump faults such as mechanical seal defects and impeller defects. CNNs suffer from the vanishing gradient problem as the number of network layers increases. However, densely connected convolutional networks (DenseNets), with their densely connected structures, can enhance feature propagation and alleviate the vanishing gradient problem. Xiong et al. used wavelet packet transform and dynamically weighted DenseNets to implement fault diagnosis for variable-speed planetary gearboxes. These fault diagnosis methods do not mitigate the problem of neural network overfitting.

[0005] Non-contact measurement fault diagnosis based on acoustic signals: Wang Xin et al. constructed a one-dimensional convolutional neural network to diagnose air conditioner motor faults using acoustic signals. They also used the t-distributed neighborhood embedding algorithm to visualize the performance of the one-dimensional convolutional neural network. Li Shaobo et al. proposed a gear fault diagnosis method based on a convolutional neural network and multi-channel acoustic signals. This method fuses the features of acoustic signals from multiple sensors to diagnose faults in multi-stage transmission gears. These diagnostic methods have not yet been applied in the field of centrifugal pumps. Summary of the Invention

[0006] In view of the shortcomings of the existing technology mentioned above, the present invention solves the problem of not being able to retain the time information of fault characteristics by converting one-dimensional time series into two-dimensional images, which helps to improve the fault recognition rate. It also uses the Bayesian optimization algorithm to obtain the hyperparameters of the neural network, which can effectively reduce the dependence on expert experience. The optimized neural network has a faster convergence speed and fault recognition accuracy, and can also alleviate the gradient vanishing problem of the neural network and suppress overfitting.

[0007] To achieve the above-mentioned purpose and other related purposes, the present invention provides a fault diagnosis method for the acoustic radiation signal of a centrifugal pump, including: obtaining a first acoustic radiation signal under different working conditions; filtering the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal; using the Gram angle field to convert the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set; using the sample set to train an improved densely connected convolutional network, and during the training process using a Bayesian optimization algorithm to optimize the hyperparameter combination of the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model; and using the acoustic radiation signal fault diagnosis model to diagnose centrifugal pump faults.

[0008] In an optional embodiment of the present invention, the step of using the Gram angular field to convert the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set includes: normalizing the one-dimensional time series data of the second acoustic radiation signal; encoding the normalized one-dimensional time series data using polar coordinates; calculating the Gram and angular field using the encoded one-dimensional time series data; and converting the one-dimensional time series into a two-dimensional GASF image as a sample set using the Gram and angular field.

[0009] In an optional embodiment of the present invention, encoding the normalized one-dimensional time series data using polar coordinates is implemented by the following formula:

[0010]

[0011] Among them, θ represents or The encoding of r represents the time t i The encoding of , i = 1, 2, ..., n, θ i Represents the polar angle of polar coordinates, r i represents the polar diameter of the polar coordinate system, and N represents the constant factor that regularizes the span of the polar coordinate system.

[0012] In an optional embodiment of the present invention, the calculation of the Gram and angular fields using the one-dimensional time series data after encoding mapping is achieved by the following formula:

[0013]

[0014] where GASF represents the Gram sum angle field, cosθ represents the cosine of the angle θ, and θ∈[0,π].

[0015] In an optional embodiment of the present invention, the improved densely connected convolutional network is trained using the sample set, and the hyperparameter combination of the improved densely connected convolutional network is optimized using the Bayesian optimization algorithm during the training process to obtain the sound radiation signal fault diagnosis model. The steps include: improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network; optimizing the hyperparameter combination of the initial learning rate, random gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network through the Bayesian optimization algorithm to obtain the optimal hyperparameter combination of the initial learning rate, random gradient descent momentum and L2 regularization strength; inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain the sound radiation signal fault diagnosis model.

[0016] In an optional embodiment of the present invention, the step of improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network includes: adding a densely connected layer with a 5×5 convolution kernel to each densely connected layer of the Dense block in the network parameters of the densely connected convolutional network; adding a dropout layer between the global average pooling layer and the fully connected layer in the network parameters of the densely connected convolutional network; and obtaining an improved densely connected convolutional network by adding a densely connected layer with a 5×5 convolution kernel and adding a dropout layer.

[0017] In an optional embodiment of the present invention, the step of optimizing the hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network by using the Bayesian optimization algorithm to obtain the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength includes: determining the preset optimization times of the Bayesian optimization algorithm, and taking the classification error rate of the validation set in the sample set as the objective function; selecting a first probabilistic proxy model and an acquisition function according to the Bayesian optimization algorithm; initializing the hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network by using the first probabilistic proxy model to obtain a first hyperparameter combination; and optimizing the hyperparameter combination according to the first hyperparameter. The first probability proxy model is updated by a combination of several parameters to obtain a second probability proxy model; the hyperparameter combination is input into the improved densely connected convolutional network; the improved densely connected convolutional network is trained using the training set in the sample set, and the validation set in the sample set is predicted using the trained improved densely connected convolutional network to obtain the classification error of the validation set, i.e., the function value of the objective function; the second hyperparameter combination is obtained by maximizing the acquisition function, and then the step of optimizing the first probability proxy model according to the first hyperparameter combination to obtain the second probability proxy model is returned to, until the preset number of optimizations is reached, and the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum, and L2 regularization strength is found according to the minimum objective function value. It should be noted that after the acquisition function determines the t-1th hyperparameter combination, the probability proxy model is updated.

[0018] In an optional embodiment of the present invention, the step of inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain the sound radiation signal fault diagnosis model includes: taking the classification error rate of the verification set in the sample set as the objective function; inputting the hyperparameter combination determined each time by the acquisition function into the improved densely connected convolutional network; training the improved densely connected convolutional network using the training set in the sample set to obtain an improved densely connected convolutional network that has completed training; predicting the verification set in the sample set by the improved densely connected convolutional network that has completed training to obtain the function value of the objective function; and evaluating the improved densely connected convolutional network that has completed training using the function value of the objective function to obtain the sound radiation signal fault diagnosis model.

[0019] To achieve the above-mentioned objectives and other related objectives, the present invention also provides a product including: an acoustic radiation signal fault diagnosis model; a target product; the acoustic radiation signal fault diagnosis model includes an acoustic radiation signal fault diagnosis method, and the acoustic radiation signal fault diagnosis method includes: obtaining a first acoustic radiation signal under different working conditions; filtering the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal; using the Gram angle field to convert the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set; using the sample set to train an improved densely connected convolutional network, and during the training process using a Bayesian optimization algorithm to optimize the hyperparameter combination of the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model; placing the acoustic radiation signal fault diagnosis model into the target product to complete the fault diagnosis of the centrifugal pump.

[0020] In order to achieve the above-mentioned purpose and other related purposes, the present invention also provides a fault diagnosis system for centrifugal pump acoustic radiation signals, including: an acquisition module, which acquires a first acoustic radiation signal under different working conditions; a filtering module, which filters the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal; a conversion module, which uses the Gram angle field to convert the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set; a training module, which uses the sample set to train an improved densely connected convolutional network, and uses a Bayesian optimization algorithm to optimize the hyperparameter combination of the improved densely connected convolutional network during the training process to obtain an acoustic radiation signal fault diagnosis model; and a diagnosis module, which uses the acoustic radiation signal fault diagnosis model to diagnose centrifugal pump faults.

[0021] The technical effect of the present invention is to provide a fault diagnosis method for the acoustic radiation signal of a centrifugal pump. By converting a one-dimensional time series into a two-dimensional image, the present invention can retain the time information of the fault characteristics, which helps to improve the fault recognition rate. At the same time, the Bayesian optimization algorithm is used to obtain the hyperparameters of the neural network, which can effectively reduce the dependence on expert experience. The optimized neural network has a faster convergence speed and higher fault recognition accuracy. It also uses multi-scale convolution to extract image features and enhance feature reuse, which can effectively extract the characteristic information of the fault and improve the fault recognition rate. A dropout layer is added to the densely connected convolutional network to randomly discard 50% of the hidden layer nodes to prevent overfitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is an application scenario diagram of a centrifugal pump acoustic radiation signal fault diagnosis method proposed by the present invention;

[0023] Figure 2 This is a flow chart of a method for fault diagnosis of centrifugal pump acoustic radiation signals proposed by the present invention;

[0024] Figure 3 This is a specific flow chart of the conversion into a two-dimensional GASF image proposed by the present invention;

[0025] Figure 4 This is a flow chart of the acoustic radiation signal fault diagnosis model proposed by the present invention;

[0026] Figure 5 A specific flow chart of the improved densely connected convolutional network proposed in the present invention;

[0027] Figure 6 A specific flow chart for obtaining the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum, and L2 regularization strength proposed by the present invention;

[0028] Figure 7 This is a specific flow chart of the acoustic radiation signal fault diagnosis model proposed by the present invention;

[0029] Figure 8 A diagram showing the process of obtaining a two-dimensional image proposed by the present invention;

[0030] Figure 9 This is the structural diagram of the densely connected convolutional network proposed in the present invention;

[0031] Figure 10 This is the improved Dense block structure diagram proposed by the present invention;

[0032] Figure 11 This is a training progress curve of the Bayesian optimization improved densely connected convolutional network diagnostic model proposed in the present invention;

[0033] Figure 12 The confusion matrix of the Bayesian optimization-improved densely connected convolutional network diagnostic model proposed in the present invention;

[0034] Figure 13 This is a training progress curve of the diagnostic model based on convolutional neural network proposed in the present invention;

[0035] Figure 14 The confusion matrix of the convolutional neural network-based diagnostic model proposed in this invention;

[0036] Figure 15 This is a training progress curve of the densely connected convolutional network diagnostic model proposed in the present invention;

[0037] Figure 16 The confusion matrix of the densely connected convolutional network diagnostic model proposed in this invention;

[0038] Figure 17 This is a training progress curve of the improved densely connected convolutional network diagnostic model proposed in the present invention;

[0039] Figure 18 The confusion matrix of the improved densely connected convolutional network diagnostic model proposed in this invention;

[0040] Figure 19 This is a functional module diagram of the centrifugal pump acoustic radiation signal fault diagnosis system proposed by the present invention;

[0041] Figure 20 This is a structural block diagram of the electronic device proposed by the present invention. DETAILED DESCRIPTION

[0042] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0043] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0044] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0045] Figure 1The application scenario diagram of the fault diagnosis method of the centrifugal pump acoustic radiation signal provided by the embodiment of the present invention, the present invention obtains the first acoustic radiation signal through an acoustic sensor, obtains the second acoustic radiation signal after filtering by a filter, converts the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image, divides the two-dimensional GASF image into a training set and a test set according to a certain ratio, builds a densely connected convolutional network, and obtains an improved densely connected convolutional network by changing the structure of the densely connected convolutional network. Then, the Bayesian optimization algorithm is used to find the best hyperparameter combination, and the improved densely connected convolutional network trained by the best hyperparameter combination is obtained. The improved densely connected convolutional network that has completed training is used as the acoustic radiation signal fault diagnosis model. In other application scenarios, the fault diagnosis for centrifugal pumps can be set according to actual conditions, and the embodiments of the present invention are not limited to this.

[0046] The electronic device may be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), an interactive network television (IPTV), a smart wearable device, etc.

[0047] The electronic device may further include a network device and / or a user device, wherein the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0048] The network where the electronic device is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.

[0049] Figure 2 This is a flow chart of a centrifugal pump acoustic radiation signal fault diagnosis method provided by an embodiment of the present invention. It should be noted that the present invention is applied to non-contact fault signal pickup. This method can be applied to Figure 1 The implementation environment shown is a schematic diagram. It should be understood that the method can also be applied to other exemplary implementation environments and specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0050] like Figure 2 As shown, a fault diagnosis method of a centrifugal pump acoustic radiation signal of this embodiment at least includes:

[0051] Step S10: Acquire first acoustic radiation signals under different working conditions. It should be noted that the different working conditions here include acoustic radiation signals under four working conditions: normal working conditions, cavitation, misalignment between the motor and the centrifugal pump shaft, and loose bolts of the centrifugal pump.

[0052] Step S20: filtering the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal.

[0053] Step S30: Using the Gram angular field, the one-dimensional time series of the second acoustic radiation signal is converted into a two-dimensional GASF image as a sample set. It should be noted that the Gram angular field uses the Gram matrix to calculate the linear correlation between a set of vectors, which can convert the one-dimensional time series into a two-dimensional image while preserving the temporal feature information.

[0054] Step S40: Using the sample set to train the improved densely connected convolutional network, during the training process, the Bayesian optimization algorithm is used to optimize the hyperparameter combination of the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model. It should be noted that the sample set includes a training set and a validation set.

[0055] Furthermore, the Bayesian optimization algorithm is an algorithm that searches for the global extreme value of the objective function within a bounded domain based on the Bayesian principle. It is suitable for optimizing black box problems and is often used for parameter optimization in machine learning. The essence of the Bayesian optimization process is to continuously update and optimize the unknown function based on the sampling points, and then find the best evaluation point and the optimal observation value.

[0056] Step S50: diagnose centrifugal pump faults using the acoustic radiation signal fault diagnosis model.

[0057] like Figure 3 As shown, the specific steps of converting the image into a two-dimensional GASF image provided in this embodiment include:

[0058] Step S31: normalizing the one-dimensional time series data of the second acoustic radiation signal. It should be noted that the normalizing of the one-dimensional time series data of the second acoustic radiation signal is achieved by the following formula:

[0059]

[0060]

[0061] Where X={x2,x2,…,x n} represents the time series data of length n, x i ∈X, Indicates that X is normalized to [-1,1], Indicates that X is normalized to [0,1], i = 1, 2, …, n.

[0062] Step S32: Encode the normalized one-dimensional time series data using polar coordinates. It should be noted that encoding the normalized one-dimensional time series data using polar coordinates is achieved by the following formula:

[0063]

[0064] Among them, θ represents or The encoding of r represents the time t i The encoding of , i = 1, 2, ..., n, θ i Represents the polar angle of polar coordinates, r i represents the polar diameter of the polar coordinate system, and N represents the constant factor that regularizes the span of the polar coordinate system.

[0065] Specifically, when θ∈[0,π], cosθ is monotonic, meaning that for a given time series, there is exactly one mapping to it in polar coordinates. Polar coordinates also preserve absolute temporal relationships, but data rescaled at different time intervals have different angular bounds, such that a cosine value of [0,1] corresponds to [0,π / 2].

[0066] Step S33: Calculate the Gram and angular field using the encoded one-dimensional time series data. It should be noted that the calculation of the Gram and angular field is achieved by the following formula:

[0067]

[0068] Where GAGF represents the Gram sum angle field, and cosθ represents the cosine of the angle θ ∈ [0,π].

[0069] Step S34: using Gram and angular fields to convert the one-dimensional time series into a two-dimensional GASF image as a sample set. Figure 8 As shown, (a) is the time domain diagram of the original signal, (b) is the mapping converted to the polar coordinate system, and (c) is the converted GASF. The time of the signal represented by the Gram angular field image gradually increases from the upper left corner to the lower right corner. Furthermore, the sample set is a two-dimensional GASF image.

[0070] like Figure 4 As shown, the specific steps of obtaining the acoustic radiation signal fault diagnosis model provided in this embodiment include:

[0071] Step S41: Build a densely connected convolutional network.

[0072] Step S42: Improve the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network. It should be noted that the network structure includes a Dense block, a global average pooling layer, and a fully connected layer.

[0073] Step S43: Optimizing the hyperparameter combination of the initial learning rate, stochastic gradient descent momentum, and L2 regularization strength of the improved densely connected convolutional network using a Bayesian optimization algorithm to obtain an optimal hyperparameter combination of the initial learning rate, stochastic gradient descent momentum, and L2 regularization strength. It should be noted that the hyperparameter combination includes the initial learning rate, stochastic gradient descent momentum, and L2 regularization strength of the densely connected convolutional network.

[0074] Step S44: Inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model. It should be noted that the trained improved densely connected convolutional network is the acoustic radiation signal fault diagnosis model.

[0075] like Figure 5 As shown, the specific steps of the improved densely connected convolutional network provided in this embodiment include:

[0076] Step S421: Add a dense connection layer with a 5×5 convolution kernel to each dense connection layer of the Dense block in the network structure of the dense connection convolution network. It should be noted that the dense connection convolution network is mainly composed of Dense blocks and transition layers, such as Figure 9 The following diagram shows the network structure of a densely connected convolutional network, which contains three Dense blocks. In the figure, BN is the batch normalization layer, Relu is the activation layer, Covl is the convolution layer, AvgPool is the pooling layer, and Transitionlayer is the transition layer.

[0077] Specifically, the dense connection layer consists of a batch normalization layer, an activation layer, and a convolutional layer. Multiple dense connection layers are connected in series to form a Dense block. Within the Dense block, the size of the output feature map of each network layer is the same. The outputs of the first n-1 dense connection layers are spliced ​​in parallel along the channel dimension to form the input of the nth dense connection layer, forming a dense connection within the Dense block.

[0078] Specifically, the output of the nth densely connected layer is x n :

[0079] x n =H n ([x1,x2,…,x n-1 ])

[0080] Among them, H nrepresents the comprehensive transformation function of n layers, [x1,x2,…,x n-1 ] represents the parallel connection of the first n-1 layers of feature maps.

[0081] Dense connections exist only within Dense blocks. This connection method ensures feature reuse and reduces the amount of computation per layer. Each Dense block is connected through a transition layer.

[0082] Furthermore, the transition layer is composed of a batch normalization layer, an activation layer, a convolutional layer, and a pooling layer. Due to the dense connections within the Dense block, the input channel dimension of each layer within the block increases with the number of layers. To control the number of channels, a 1×1 convolutional layer is used in the transition layer to compress the feature map dimension. A 2×2 average pooling layer (with a stride of 2) in the transition layer can reduce the size of the feature map. Together, these two network layers reduce the computational complexity of the neural network.

[0083] In a specific embodiment, if Figure 10 As shown in the figure, a dense connection layer with a 5×5 convolution kernel is added to each dense connection layer to achieve multi-scale extraction and reuse of image features and prevent overfitting.

[0084] Step S422: Add a dropout layer between the global average pooling layer and the fully connected layer in the network structure of the densely connected convolutional network. It should be noted that the dropout layer is added between the global average pooling layer and the fully connected layer in the network parameters of the densely connected convolutional network to prevent overfitting. The dropout layer can randomly set the hidden layer activation value to 0 according to a specified ratio, thereby invalidating the hidden layer nodes. For each batch of densely connected convolutional network training, since the dropout layer ignores the randomness of the hidden layer nodes, the invalid hidden layer nodes are not exactly the same, and the densely connected convolutional network trained each time is equivalent to a new one.

[0085] Specifically, the network parameters of the densely connected convolutional network are optimized and improved, and the initial image input size is 64×64×3, as shown in Table 1:

[0086] Table 1

[0087]

[0088] Step S423: By adding a densely connected layer with a 5×5 convolution kernel and a dropout layer, an improved densely connected convolutional network is obtained. It should be noted that the structure of the densely connected convolutional network is improved and the network parameters are optimized.

[0089] like Figure 6As shown, the specific steps provided in this embodiment for obtaining the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum, and L2 regularization strength include:

[0090] Step S431: Determine a preset number of optimizations for the Bayesian optimization algorithm, and use the classification error rate of the validation set in the sample set as the objective function. It should be noted that Bayesian optimization is an algorithm that searches for the global extremum of an objective function within a bounded domain based on the Bayesian principle. It is suitable for optimizing black-box problems and is commonly used for parameter optimization in machine learning. The Bayesian optimization algorithm includes a probabilistic surrogate model and an acquisition function.

[0091] Step S432: Select a first probabilistic proxy model and an acquisition function based on the Bayesian optimization algorithm. It should be noted that the first probabilistic proxy model is an assumed function based on the objective function, including a prior probability distribution and a posterior probability distribution. Specifically, a Gaussian process is used as the probabilistic proxy model. Furthermore, the objective function is the classification error rate of the validation set in the sample set.

[0092] The acquisition function is Expected Improvement, which is used to determine the next set of hyperparameter combinations.

[0093] In a specific embodiment, the Gaussian process is a non-parametric model, which is a set of any finite random variables with a joint Gaussian distribution, consisting of a mean function and a covariance function. The expression of the Gaussian process is as follows:

[0094]

[0095] Where m(x) represents the mean function, k(x, x′) represents the covariance function, f(x) represents the objective function, GP represents Gaussian distribution, and E represents expectation.

[0096] The mean function m(x) is usually set to 0, and the prior distribution p(f|X,θ) of the Gaussian process is as follows:

[0097] p(f|X,θ)~N(0,K)

[0098] Where X represents the set of hyperparameter combinations, f represents the set of objective function values, θ is a parameter, K represents the covariance matrix composed of k(x,x′), and N represents the normal distribution.

[0099] Step S433: Initialize the vector of the hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network through the first probabilistic proxy model to obtain a first hyperparameter combination.

[0100] Step S434: updating the first probabilistic proxy model according to the first hyperparameter combination to obtain a second probabilistic proxy model.

[0101] Step S435: Input the hyperparameter combination into the improved densely connected convolutional network, use the training set in the sample set to train the improved densely connected convolutional network, use the trained improved densely connected convolutional network to predict the validation set in the sample set, and obtain the classification error of the validation set, that is, the function value of the objective function.

[0102] Step S436: After maximizing the acquisition function to obtain the second hyperparameter combination, the process returns to the step of optimizing the first probabilistic proxy model according to the first hyperparameter combination to obtain the second probabilistic proxy model, until the preset number of optimizations is reached, and the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum, and L2 regularization strength is found according to the minimum objective function value.

[0103] Specifically, the acquisition function is used to determine the next hyperparameter combination from the parameter space. The expression of the acquisition function for determining the next hyperparameter combination is as follows:

[0104]

[0105]

[0106] Among them, f(x * ) represents the optimal function value of the current objective function, μ(x) represents the mean, σ(x) represents the standard deviation, α EI (x) represents the acquisition function.

[0107] Maximize the acquisition function to get the next hyperparameter combination, which is achieved by the following formula:

[0108] x t+1 =argmax α EI (x; D 1:t )

[0109] Among them, x t+1 is the t+1th hyperparameter combination, α EI (x) represents the acquisition function, D 1:t Represents the dataset D 1:t ={(x1,f1),…,(x1,f t )},(x t ,f t ) indicates the data obtained after the tth iteration of the Bayesian optimization algorithm, x t represents the tth hyperparameter combination, f t Represents the function value of the t-th objective function.

[0110] When the next hyperparameter combination is determined, the probability proxy model is updated, and the posterior probability distribution p of the objective function at this time is as follows:

[0111] p(f t+1 |D 1:t ,x t+1 )~GP(μ,σ 2 )

[0112]

[0113]

[0114] k t+1 =[k(x t+1 ,x1),…,k(x t+1 ,x t+1 )]

[0115] Among them, x t+1 is the t+1th hyperparameter combination, μ(x) represents the mean, σ(x) represents the standard deviation, and f 1:t+1 The vector representing the function values ​​of the first t+1 objective functions, k t+1 represents the vector of the first t+1 covariance functions, represents k t+1 The transpose of K -1 represents the inverse matrix of the covariance matrix, k(x i ,x j ) represents the value of x i and x j The covariance function of .

[0116] After the probabilistic proxy model is updated, the hyperparameter combination is input into the improved densely connected convolutional network. The neural network is trained using the training set in the sample set and cross-validated using the validation set to obtain the function value of the objective function. The acquisition function is then returned to the step of determining the next evaluation point until the preset number of optimizations is reached. Finally, the optimal hyperparameter combination corresponding to the optimal objective function value is found among all hyperparameter combinations. It should be noted that after the t-th hyperparameter combination is determined, the probabilistic proxy model for the t-1th time is updated.

[0117] like Figure 7 As shown, the specific steps of obtaining the acoustic radiation signal fault diagnosis model provided in this embodiment include:

[0118] Step S441: taking the classification error rate of the validation set in the sample set as the objective function.

[0119] Step S442: Inputting the hyperparameter combination determined by the acquisition function each time into the improved densely connected convolutional network. It should be noted that each time the acquisition function determines the next hyperparameter combination, each hyperparameter combination will be input into the improved densely connected convolutional network.

[0120] Step S443: The improved densely connected convolutional network is trained using the training set in the sample set to obtain a trained improved densely connected convolutional network. It should be noted that after each hyperparameter combination is input into the improved densely connected convolutional network, the improved densely connected convolutional network is trained.

[0121] Step S444: Predicting the validation set in the sample set using the trained improved densely connected convolutional network to obtain the function value of the objective function. The post-trained improved densely connected convolutional network means that each time a hyperparameter combination is determined, the corresponding training is completed. When the next training begins, a new round of training is performed on the initial improved densely connected convolutional network. Each trained improved densely connected convolutional network predicts the validation set in the sample set to obtain the function value of the objective function.

[0122] Step S445: Evaluate the trained improved densely connected convolutional network using the function value of the objective function to obtain an acoustic radiation signal fault diagnosis model. It should be noted that the improved densely connected convolutional network is evaluated based on the function value of the objective function obtained during each training session to assess whether the current hyperparameter combination is the current optimal combination and whether the improved densely connected convolutional network is optimal. If the improved densely connected convolutional network is optimal, the optimal improved densely connected convolutional network is used as the acoustic radiation signal fault diagnosis model.

[0123] like Figure 11-18 As shown, the technical solution of the present invention is described below with reference to specific embodiments:

[0124] It should be noted that the acoustic radiation signal of the HTE-1 centrifugal pump body is collected with a sampling frequency of 51200 Hz and a motor speed stabilized at 2950 r / min. The acoustic radiation signals of the centrifugal pump are collected under four working conditions: normal, cavitation, misalignment between the motor and the centrifugal pump shaft (referred to as shaft misalignment), and loose bolts of the centrifugal pump.

[0125] The data sample length was set to 1024. Based on the time series visualization method, the one-dimensional data samples for the four operating conditions were first normalized to [-1, 1] and then converted into two-dimensional GASF images using the Gram angle field. The two-dimensional image data was then constructed into Dataset 1 and Dataset 2. Dataset 1 contains 500 samples for each operating condition, while Dataset 2 contains 100 samples for each operating condition. As shown in Table 2, one-hot encoding was used to construct four fault labels for the centrifugal pump:

[0126] Table 2

[0127]

[0128] Specifically, the model is trained using a cross-validation method. Dataset 1 is randomly divided into a training set and a validation set in a ratio of 7:3. Dataset 2 is used as a test set to test the trained neural network model.

[0129] The number of Bayesian optimization training iterations was set to 30, the SGDM optimizer was used, the minimum batch size was 128, and the optimized neural network hyperparameter value ranges are shown in Table 3. The results of Bayesian optimization to improve the hyperparameters of the densely connected convolutional network are shown in Table 4.

[0130] Table 3

[0131]

[0132] Table 4

[0133]

[0134] According to Table 4, the neural network model obtained in the 19th time is the best, and the best hyperparameter combination is: the initial learning rate is 0.0057624, the momentum is 0.82065, and the L2 regularization strength is 0.0085661.

[0135] like Figure 11-12 As shown, Figure 11 This is a progress curve of the hyperparameter combination training improved densely connected convolutional network after Bayesian optimization. It can be seen from the figure that after 22 iterations, the training accuracy of the neural network tends to be stable. The diagnostic model is tested using Dataset 2. The confusion matrix of the test results is as follows Figure 12 As shown in the figure, the x-axis and y-axis of the confusion matrix represent the predicted class and the true class, respectively. The column summary on the right represents the accuracy of each working condition, and the row summary on the bottom represents the recall rate of each working condition. From the column summary, we can calculate that the average recognition accuracy of the model fault is 99.5%.

[0136] Convolutional neural network, densely connected convolutional neural network, and improved densely connected convolutional neural network are used as comparison models. It should be noted that the three models are trained using Dataset 1 in a cross-validation manner. Dataset 1 is randomly divided into training set and validation set in a ratio of 7:3. The trained models are tested using Dataset 2. The hyperparameters of the three comparison models are set as follows: initial learning rate 0.001, momentum 0.8, L2 regularization strength 1×10 -6 .

[0137] like Figure 13-18 As shown, Figure 13 This is a training progress curve of the diagnostic model based on convolutional neural network. Figure 14 is the confusion matrix of the test results of the diagnostic model based on convolutional neural network, Figure 15 This is a training progress curve of the densely connected convolutional network diagnostic model. Figure 16 is the confusion matrix of the test results based on the densely connected convolutional network diagnostic model, Figure 17 To improve the training progress graph of the densely connected convolutional network diagnostic model, Figure 18 Confusion matrix of test results for improving the diagnostic model for densely connected convolutional networks.

[0138] Specifically, the average recognition accuracy of the validation set and test set of the four models is shown in Table 5:

[0139] Table 5

[0140]

[0141] According to the model training and test results, the Bayesian optimization improved densely connected convolutional network has the best performance among the four diagnostic models, fast convergence speed, and the highest fault recognition rate for the test set; Figure 13-14 From Table 5, we can see that the convolutional neural network has a slow training speed and a low recognition rate for working condition 2 and working condition 4, which are 85% and 81% respectively. Its average recognition accuracy rate of the test set is the lowest. Compared with the convolutional neural network, the densely connected convolutional network has a faster training speed. Although its recognition rate for working condition 1 and working condition 3 is low, its recognition rate for the other two working conditions has improved. Moreover, its average recognition rate of the test set is higher than that of the convolutional neural network. Figure 15-18 It can be seen that the training accuracy of the improved densely connected convolutional network is stable after 35 iterations, while the training accuracy of the densely connected convolutional network fluctuates after 40 iterations. The training accuracy of the improved densely connected convolutional network converges at a faster rate than that of the densely connected convolutional network, and the recognition rate of each working condition is 95% or above. In addition, compared with the densely connected convolutional network, the average recognition rate of the test set fault of the improved densely connected convolutional network is increased by 5.25%, indicating the effectiveness of multi-scale convolution kernel in extracting two-dimensional image features; after Bayesian optimization of the improved densely connected convolutional network, the training speed of the neural network is improved, and the recognition rate of each working condition reaches 97% or above.

[0142] like Figure 19As shown, the present invention also provides a fault diagnosis system for centrifugal pump sound radiation signals, including: an acquisition module 191, a filtering module 192, a conversion module 193, a training module 194, and a diagnosis module 195. The acquisition module 191 is used to acquire a first sound radiation signal under different working conditions; the filtering module 192 is used to filter the first sound radiation signal through a non-recursive filter to obtain a second sound radiation signal; the conversion module 193 is used to use the Gram angle field to convert the one-dimensional time series of the second sound radiation signal into a two-dimensional GASF image as a sample set; the training module 194 is used to use the sample set to train an improved densely connected convolutional network, and during the training process, the Bayesian optimization algorithm is used to optimize the hyperparameter combination of the improved densely connected convolutional network to obtain a sound radiation signal fault diagnosis model; the diagnosis module 195 is used to use the sound radiation signal fault diagnosis model to diagnose centrifugal pump faults.

[0143] It should be noted that the above embodiments provide Figure 19 The fault diagnosis system for the acoustic radiation signal of a centrifugal pump shown in the embodiment is based on the same concept as the fault diagnosis method for the acoustic radiation signal of a centrifugal pump provided in the above embodiment. The specific manner in which the various modules and units perform their operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the fault diagnosis system for the acoustic radiation signal of a centrifugal pump provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the device into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0144] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the fault diagnosis method of the centrifugal pump acoustic radiation signal provided in the above-mentioned embodiments.

[0145] Figure 20 FIG1 shows a schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present invention. Figure 20 The computer system 2000 of the electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0146] like Figure 20As shown, computer system 2000 includes a central processing unit (CPU) 2001, which can perform various appropriate actions and processes according to programs stored in read-only memory (ROM) 2002 or programs loaded from storage unit 2006 into random access memory (RAM) 2003, such as executing the methods described in the above embodiments. Various programs and data required for system operation are also stored in RAM 2003. CPU 2001, ROM 2002, and RAM 2003 are connected to each other via bus 2004. Input / output (I / O) interface 2005 is also connected to bus 2004.

[0147] The following components are connected to the I / O interface 2005: an input section 2006 including a keyboard, a mouse, and the like; an output section 2007 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 2006 including a hard disk; and a communication section 2008 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 2008 performs communication processing via a network such as the Internet. A drive 2010 is also connected to the I / O interface 2005 as needed. Removable media 2011, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 2010 as needed, so that computer programs read therefrom can be installed into the storage section 2006 as needed.

[0148] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a computer-readable medium, the computer program including a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 2008 and / or installed from a removable medium 2011. When the computer program is executed by the central processing unit (CPU) 2001, the various functions defined in the system of the present invention are performed.

[0149] It should be noted that the computer-readable medium shown in the embodiments of the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium may, for example, be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable computer program. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0151] The units involved in the embodiments of the present invention may be implemented in software or hardware, and the units described may also be provided in a processor. In some cases, the names of these units do not limit the units themselves.

[0152] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When executed by a computer processor, the computer program causes the computer to perform the aforementioned method for diagnosing a centrifugal pump acoustic radiation signal fault. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist independently and not be incorporated into the electronic device.

[0153] Another aspect of the present invention provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the centrifugal pump acoustic radiation signal fault diagnosis method provided in each of the above embodiments.

[0154] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, any equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A method for fault diagnosis of centrifugal pump acoustic radiation signals, characterized in that: include: Obtain the first acoustic radiation signal under different working conditions; Filtering the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal; using a Gram angle field to convert the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set; Using the sample set to train an improved densely connected convolutional network, and during the training process using a Bayesian optimization algorithm to optimize the hyperparameter combination of the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model; Performing centrifugal pump fault diagnosis using the acoustic radiation signal fault diagnosis model; The step of training the improved densely connected convolutional network using the sample set and optimizing the hyperparameter combination of the improved densely connected convolutional network using the Bayesian optimization algorithm during the training process to obtain the acoustic radiation signal fault diagnosis model includes: Build a densely connected convolutional network; Improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network; The Bayesian optimization algorithm is used to optimize the hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network to obtain the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength; Inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model; The step of improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network includes: Adding a dense connection layer with a 5×5 convolution kernel to each dense connection layer of the Dense block in the network parameters of the densely connected convolutional network; Adding a dropout layer between the global average pooling layer and the fully connected layer in the network parameters of the densely connected convolutional network; By adding a densely connected layer with a 5×5 convolution kernel and a dropout layer, an improved densely connected convolutional network is obtained.

2. The fault diagnosis method of the centrifugal pump acoustic radiation signal according to claim 1, characterized in that: The step of converting the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set by using the Gram angle field comprises: performing normalization processing on the one-dimensional time series data of the second acoustic radiation signal; Encoding the normalized one-dimensional time series data using polar coordinates; Calculating Gram and angular fields using the encoded one-dimensional time series data; The one-dimensional time series is converted into a two-dimensional GASF image using Gram and angular fields as a sample set.

3. The fault diagnosis method of the centrifugal pump acoustic radiation signal according to claim 2, characterized in that: The encoding of the normalized one-dimensional time series data using polar coordinates is achieved by the following formula: Among them, θ represents or The encoding of r represents time The encoding of , i=1,2,…,n, represents the polar angle of polar coordinates, represents the polar diameter of polar coordinates, A constant factor representing the span of the regularized polar coordinate system.

4. The fault diagnosis method of the centrifugal pump acoustic radiation signal according to claim 2, characterized in that: The calculation of the Gram and angular fields using the encoded one-dimensional time series data is achieved by the following formula: in, represents the Gram and angle fields, θ represents the cosine of the angle θ, θ∈[0,π].

5. The fault diagnosis method of the centrifugal pump acoustic radiation signal according to claim 1, characterized in that: The step of optimizing the hyperparameter combination of the initial learning rate, stochastic gradient descent momentum, and L2 regularization strength of the improved densely connected convolutional network by the Bayesian optimization algorithm to obtain the optimal hyperparameter combination of the initial learning rate, stochastic gradient descent momentum, and L2 regularization strength includes: Determining a preset number of optimizations for the Bayesian optimization algorithm, and using the classification error rate of the validation set in the sample set as an objective function; selecting a first probabilistic proxy model and an acquisition function according to a Bayesian optimization algorithm; Initializing a hyperparameter combination of an initial learning rate, a stochastic gradient descent momentum, and an L2 regularization strength of the improved densely connected convolutional network using the first probabilistic proxy model to obtain a first hyperparameter combination; Updating the first probabilistic proxy model according to the first hyperparameter combination to obtain a second probabilistic proxy model; Inputting the hyperparameter combination into the improved densely connected convolutional network, training the improved densely connected convolutional network using the training set in the sample set, and using the trained improved densely connected convolutional network to predict the validation set in the sample set, to obtain the classification error of the validation set, i.e., the function value of the objective function; By maximizing the acquisition function to obtain the second hyperparameter combination, the method returns to the step of optimizing the first probabilistic proxy model according to the first hyperparameter combination to obtain the second probabilistic proxy model, until the preset number of optimizations is reached, and the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum, and L2 regularization strength is found according to the objective function value.

6. The fault diagnosis method of the centrifugal pump acoustic radiation signal according to claim 5, characterized in that: The step of inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model includes: The classification error rate of the validation set in the sample set is used as the objective function; Inputting the hyperparameter combination determined each time by the acquisition function into the improved densely connected convolutional network; Training the improved densely connected convolutional network using the training set in the sample set to obtain a trained improved densely connected convolutional network; Predicting the validation set in the sample set by using the trained improved densely connected convolutional network to obtain the function value of the objective function; The function value of the objective function is used to evaluate the improved densely connected convolutional network that has completed training to obtain an acoustic radiation signal fault diagnosis model.

7. A centrifugal pump acoustic radiation signal fault diagnosis system, characterized in that: include: An acquisition module, which acquires the first acoustic radiation signal under different working conditions; a filtering module, filtering the first acoustic radiation signal through a non-recursive filter to obtain a second acoustic radiation signal; a conversion module, which converts the one-dimensional time series of the second acoustic radiation signal into a two-dimensional GASF image as a sample set by using a Gram angle field; A training module, which uses the sample set to train an improved densely connected convolutional network, and uses a Bayesian optimization algorithm to optimize the hyperparameter combination of the improved densely connected convolutional network during the training process to obtain an acoustic radiation signal fault diagnosis model; a diagnostic module for diagnosing centrifugal pump faults using the acoustic radiation signal fault diagnosis model; The step of training the improved densely connected convolutional network using the sample set and optimizing the hyperparameter combination of the improved densely connected convolutional network using the Bayesian optimization algorithm during the training process to obtain the acoustic radiation signal fault diagnosis model includes: Build a densely connected convolutional network; Improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network; The Bayesian optimization algorithm is used to optimize the hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength of the improved densely connected convolutional network to obtain the optimal hyperparameter combination of initial learning rate, stochastic gradient descent momentum and L2 regularization strength; Inputting the optimal hyperparameter combination into the improved densely connected convolutional network and training the improved densely connected convolutional network to obtain an acoustic radiation signal fault diagnosis model; The step of improving the structure of the densely connected convolutional network to obtain an improved densely connected convolutional network includes: Adding a dense connection layer with a 5×5 convolution kernel to each dense connection layer of the Dense block in the network parameters of the densely connected convolutional network; Adding a dropout layer between the global average pooling layer and the fully connected layer in the network parameters of the densely connected convolutional network; By adding a densely connected layer with a 5×5 convolution kernel and a dropout layer, an improved densely connected convolutional network is obtained.

8. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the fault diagnosis method of the centrifugal pump acoustic radiation signal as described in any one of claims 1-6.

Images