Centrifugal pump fault diagnosis method and device based on continuous wavelet transform and deep learning and electronic equipment

Abstract

The invention discloses a centrifugal pump fault diagnosis method and device based on continuous wavelet transform and deep learning and electronic equipment, and belongs to the technical field of mechanical equipment fault diagnosis. The method comprises the steps that a centrifugal pump vibration signal is collected through an acceleration sensor, after normalization, denoising and sliding window framing preprocessing are conducted, continuous wavelet transform (CWT) is conducted through a Morlet wavelet basis function, and a 280 * 280 time-frequency graph with the scale range being 1-128 is generated; a five-layer convolutional neural network (CNN) is adopted to automatically extract deep features of the time-frequency graph, and 256-dimensional feature vectors are output; retaining 95% of cumulative variance contribution rate in combination with principal component analysis (PCA), and reducing dimensions to 30 dimensions to remove redundant information; and a grey wolf optimization algorithm (GWO) is adopted to globally search an optimal parameter combination of a support vector machine (SVM), so that high-precision multi-fault classification is realized. The method can significantly improve the non-stationary signal feature extraction capability, is suitable for an industrial complex environment, and effectively guarantees the safe operation and economical efficiency of equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of centrifugal pump fault diagnosis, and particularly relates to a centrifugal pump fault diagnosis method, device and electronic device based on continuous wavelet transform (CWT) and deep learning. Background Technique

[0002] As a core device in industrial fluid mechanics, centrifugal pumps are widely used in fields such as energy power, petrochemical industry, and agricultural water conservancy. Their operating stability directly affects the reliability and safety of the production system. Most of the faults of centrifugal pumps originate from aspects such as cavitation, impeller wear, and bearing damage. When a fault occurs, it will cause abnormal operation of the entire unit and require shutdown for maintenance. If not handled in a timely manner, it may even cause personal accidents. Therefore, accurately and timely diagnosing the faults of centrifugal pumps is crucial for reducing economic losses caused by faults and ensuring the personal safety of relevant personnel.

[0003] With the rapid development of machine learning, it has been widely applied to traditional fault diagnosis, and centrifugal pump diagnosis is no exception. Among them, the most remarkable effect is the application of convolutional neural networks to centrifugal pump fault diagnosis. Due to its super strong feature extraction ability, convolutional neural networks can extract useful feature information from limited data. However, some existing models using convolutional neural networks for fault diagnosis cannot fully extract the original information features, some network layers are too deep resulting in slow operation speed, and most methods complete the model construction based on a single data set. This leads to a serious decline in accuracy when this model is used for detecting other data sets, that is, the performance of low generalization ability, and ultimately results in inaccurate diagnosis when the model is applied to fault diagnosis. Summary of the Invention

[0004] The present invention proposes a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning to solve the problems raised in the above background technique.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning, the specific steps are as follows:

[0007] S1: Obtain multi-dimensional vibration data through a triaxial acceleration sensor, construct a data set through sliding window segmentation and data augmentation, and perform normalization processing on the data set;

[0008] S2: Convert the vibration signal into a time-frequency diagram through continuous wavelet transform;

[0009] S3: Use a convolutional neural network (CNN) to extract deep features from the time-frequency diagram;

[0010] S4: Use principal component analysis (PCA) to reduce the dimension of the features;

[0011] S5: Use a support vector machine (SVM) optimized by the grey wolf optimization algorithm (GWO) to classify the features after dimension reduction and output the fault type.

[0012] Further, the step S1 specifically includes:

[0013] (1) Use an acceleration sensor to collect the three-way and base vibration signals of the centrifugal pump, covering normal states and typical faults (such as cavitation, impeller wear, bearing damage);

[0014] (2) Normalize the signals and perform overlapping sampling with a sliding window (window length 1024, step size 256) to generate a sample set;

[0015] Further, the step S2 specifically includes:

[0016] (1) Perform continuous wavelet transform on the data in the sample set obtained in step S1. The formula for continuous wavelet transform is as follows:

[0017] In the formula, W f (a,τ) is the output result of the wavelet transform, which is two-dimensional data; f(t) is the input signal, is the complex conjugate function of the wavelet basis function, a is the scale factor, and τ is the translation factor; In this patent, the Morlet wavelet is used as the basis function, and its expression is:

[0018]

[0019] where ω 0 is the center frequency, and ψ(t) is the mother wavelet function;

[0020] (2) Divide the original vibration signal x(t) into frames with a length of 1024 points and a step size of 256 points to ensure an inter-frame overlap rate of 75% to capture short-time fault impacts, and perform normalization processing on each frame of the signal;

[0021] (3) The result obtained after continuous wavelet transform contains amplitude information amp and frequency information f. Generate the time axis t according to the set sampling frequency and sample length, and draw the time-frequency diagram corresponding to each sample based on amp, f, and t;

[0022] (4) Save the time-frequency diagrams obtained in step (3) in PNG format, and convert each time-frequency diagram into a 280×280 RGB format image to meet the training requirements of the subsequent constructed convolutional neural network, where 280×280 represents the pixel size of the time-frequency diagram; Finally, divide these images into a training set and a test set at a ratio of 4:1.

[0023] Further, step S3 specifically includes:

[0024] (1) Build a convolutional neural network model based on the Python language and the PyTorch platform, including convolutional layers, pooling layers, and fully connected layers. The specific network structure: it contains 5 convolutional layers, 5 pooling layers, and 1 fully connected layer; the convolutional layers are divided into shallow convolutional layers and deep convolutional layers, the pooling layers all use the max pooling method, and the fully connected layer flattens the feature vector into a one-dimensional vector, and the fully connected layer uses the Softmax function to activate and output the probabilities of each category;

[0025] (2) The convolutional layer performs a convolutional operation on the input feature map to extract the input features, and the stacking of multiple convolutional layers can extract higher-level feature representations; the operation formula of the convolutional layer is as follows:

[0026]

[0027] In the formula, F l (x, y) is the output feature value of the l-th layer at the position (x, y), W l is the convolutional kernel of the l-th layer (with size k h *k w , such as 5×5), F l-1 is the input feature map of the previous layer, b l is the bias term, and f(·) is the activation function;

[0028] (3) All 5 pooling layers in the CNN model use the max pooling method to select the maximum value from a local area of the input feature map as the output, which is used to reduce the spatial dimension of the feature map, reduce the number of parameters and the computational complexity; the calculation formula of the max pooling method is as follows:

[0029]

[0030] In the formula, P l (x, y) is the value of the output feature map of the max pooling layer at the position (x, y), k h *k w is the pooling window size (such as 4×4), and s is the sliding step;

[0031] (4) The fully connected layer is a multi-layer perceptron in which neurons between layers are all connected, located at the end of the model, used to integrate the features extracted by the convolutional layer and the pooling layer, and output the final prediction result.

[0032] Further, step S4 specifically includes:

[0033] (1) Standardize the high-dimensional fully connected layer features obtained in step S3 to eliminate the dimensional difference, and its processing formula is:

[0034] Among them, μ is the feature mean vector, and σ is the standard deviation vector, which eliminates the dimensional difference between features:

[0035] (2) Based on the standardized data matrix Z, calculate the covariance matrix C to reveal the linear correlation between features; perform eigen-decomposition on C using singular value decomposition (SVD) to obtain the eigenvalues λ arranged in descending order of variance contribution rate i and the corresponding orthogonal eigenvectors v i ;

[0036] (3) According to the cumulative variance contribution rate criterion, select the first 30 principal components to make their cumulative contribution rate not less than 95%; construct the projection matrix W = [v 1 , v 2 , …, v 30 , and project the standardized features into the principal component space:

[0037] Y = ZW

[0038] where Y is the feature matrix after dimensionality reduction;

[0039] (4) Output the 30-dimensional feature matrix Y as the input of the SVM classifier, reducing the training time of SVM by more than 90%.

[0040] Furthermore, the specific steps of step S5 include:

[0041] (1) Input the 30-dimensional feature matrix Y ∈ R N*30 after PCA dimensionality reduction, where each row corresponds to a sample and each column corresponds to a principal component feature; N is the total number of samples, representing the number of all centrifugal pump status data; label data: L ∈ {1, 2, 3, 4}, and each label corresponds to a failure type (1 - normal, 2 - cavitation, 3 - impeller wear, 4 - bearing damage);

[0042] (2) Perform parameter initialization, set the GWO population size, the number of iterations, and the SVM parameter search range; among them, set the penalty factor C ∈ [0, 100] to balance the classification error and the model complexity; set the radial basis kernel function parameter σ ∈ [0, 100] to control the distribution of data mapped to the high-dimensional space; set the GWO algorithm parameters, the population size N = 20, the maximum number of iterations T = 30, and the convergence factor a linearly decreases from 2 to 0 to control the search range to shift from global exploration to local exploitation;

[0043] (3) Randomly divide the training set samples into 5 parts, for the current gray wolf parameters (C, σ), train the SVM with 4 parts in turn and validate with 1 part, for a total of 5 cycles; record the accuracy of each validation, and take the average of the 5 times as the fitness value of this parameter combination:

[0044] Wherein, Accuracy K is the classification accuracy of the k-th validation set

[0045] (4) In the grey wolf optimization algorithm, the wolf pack is divided into four social ranks: α, β, δ, ω; sorted by fitness value, the top 3 are α (the best), β (the second best), δ (the third best), and their parameter combinations are the current optimal solutions; each wolf adjusts its own parameters according to the positions of α, β, and δ:

[0046]

[0047] Wherein, X α , X β , X δ are the parameter combinations of the top three excellent wolves;

[0048] (5) When the maximum number of sub-iterations (30 times) is reached, the algorithm is terminated, and the global optimal parameters C * and σ * are output;

[0049] (6) The SVM model is configured with the optimal parameter combination obtained by GWO optimization, and the radial basis kernel (RBF) is selected to achieve non-linear classification; based on the one-versus-rest (OvR) strategy, independent binary-class SVMs are trained for each fault category (normal, cavitation, impeller wear, bearing damage); the 30-dimensional features of the test set samples are input, and the decision values are calculated by 4 SVM models respectively, and the category with the largest decision value is selected as the prediction result; the confidence score is generated according to the normalized distance from the sample to the classification hyperplane, and the closer the value is to 1, the more reliable the classification result is.

[0050] An electronic device of the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the centrifugal pump fault diagnosis method based on continuous wavelet transform and convolutional neural network as described above is implemented.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows: By combining continuous wavelet transform and convolutional neural network, this method can deeply mine the complex features in vibration signals. In particular, the energy feature vectors extracted by wavelet transform are used to effectively improve the accuracy of fault diagnosis. This feature extraction method can capture more subtle signal changes, making the diagnosis results more accurate and reliable. When training the CNN model, a training set covering a wide range of different fault types is used for training, enabling the model to learn rich fault features and thus demonstrating strong generalization ability in practical applications. Due to the high efficiency of the CNN model in processing large-scale data and the fast feature extraction ability of continuous wavelet transform, this method can achieve fast and real-time diagnosis of centrifugal pump faults. This is of great significance for ensuring the safe operation of centrifugal pumps and reducing losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0055] Please refer to Figure 1 , the present invention provides a technical solution:

[0056] A method for diagnosing centrifugal pump faults based on continuous wavelet transform and deep learning, the specific steps are as follows:

[0057] S1. Obtain the original signal of the centrifugal pump in the operating state. Usually, a three-axis acceleration sensor is used to collect the vibration signal of the centrifugal pump. The acceleration sensor can convert the vibration information into an electrical signal, and then normalize the collected signal; perform overlapping sampling on the normalized data to obtain data samples of a fixed length, make labels for each sample based on the fault type, and divide the labeled samples into a training sample set and a test sample set.

[0058] S2. Perform continuous wavelet transform on the training samples and test samples obtained in S1 to obtain the time-frequency diagrams of each sample; convert the time-frequency diagrams of each sample into a three-channel RGB format to adapt to the training of the subsequent constructed convolutional neural network, and form a centrifugal pump fault diagnosis data set in the form of time-frequency diagrams, which includes a training set and a test set.

[0059] S3. Build a convolutional neural network model based on the Python language and the PyTorch platform, including a convolutional layer, a pooling layer, and a fully connected layer; the convolutional layer and the pooling layer extract deep features, and the fully connected layer is used to integrate the results and provide data for the subsequent steps.

[0060] S4. Use the principal component analysis to reduce the dimension of the 256-dimensional CNN features output in step S3. Retain the principal components with 95% data variance through orthogonal transformation, remove noise and redundant information, and compress the features to 30 dimensions. While maintaining the integrity of the fault discrimination information, it greatly reduces the computational complexity and overfitting risk of the subsequent classifier.

[0061] Based on the data in S4, combine the grey wolf optimization algorithm to globally search for the optimal parameter combination of the support vector machine, build a high-precision classification model, perform multi-state recognition (normal / cavitation / impeller wear / bearing damage) on the 30-dimensional features, and output the fault type and confidence score to achieve the classification accuracy and real-time diagnostic response in a complex noise environment.

[0062] Specifically, the data processing stage of S1 includes the following steps:

[0063] (1) Use an acceleration sensor to collect the three-way and base vibration signals of the centrifugal pump, covering normal states and typical faults (such as cavitation, impeller wear, bearing damage);

[0064] (2) Normalize the signals and perform overlapping sampling with a sliding window (window length 1024, step size 256) to generate a sample set; the data after the above normalization process will fall within the interval [0,1], which makes the data closer numerically and helps the fault diagnosis model find the optimal solution faster when optimizing parameters, improving the training speed and convergence speed of the model; select the normalized vibration data samples of the centrifugal pump in normal, cavitation, impeller wear, and bearing damage states to plot the time-domain waveform diagrams.

[0065] Further, S2 specifically includes:

[0066] (1) Perform continuous wavelet transform on the data in the sample set obtained in S1. The formula for continuous wavelet transform is as follows:

[0067]

[0068] In the formula, W f(a, τ) is the output result of the wavelet transform, which is two-dimensional data; f(t) is the input signal, which is the complex conjugate function of the wavelet basis function, a is the scale factor, and τ is the translation factor; the present invention selects the complex Morlet wavelet cmor suitable for processing vibration signals as the wavelet basis function, and its expression is:

[0069]

[0070] where ω 0 is the center frequency, and ψ(t) is the mother wavelet function;

[0071] (2) Frame the original vibration signal x(t) with a length of 1024 points and a step size of 256 points, ensuring an inter-frame overlap rate of 75% to capture short-time fault shocks, and normalize each frame of the signal;

[0072] (3) The result obtained after continuous wavelet transform contains amplitude information amp and frequency information f. Generate the time axis t based on the set sampling frequency and sample length, and draw the time-frequency diagram corresponding to each sample based on amp, f, and t;

[0073] (4) Save the time-frequency diagram obtained in step (3) in PNG format, and convert each time-frequency diagram into a 280×280 RGB format picture to meet the training requirements of the subsequent constructed convolutional neural network, where 280×280 represents the pixel size of the time-frequency diagram; finally, divide these pictures into a training set and a test set in a ratio of 4:1.

[0074] Further, the specific content of S3 includes:

[0075] (1) Build a convolutional neural network model based on the Python language and the PyTorch platform, including convolutional layers, pooling layers, and fully connected layers. The specific network structure: contains 5 convolutional layers, 5 pooling layers, and 1 fully connected layer; the convolutional layers are divided into shallow convolutional layers and deep convolutional layers. The pooling layers all use the maximum pooling method. The fully connected layer flattens the feature vector into a one-dimensional vector, and the fully connected layer uses the Softmax function to activate and output the probabilities of each category;

[0076] (2) The convolutional layer performs a convolution operation on the input feature map to extract the input features. The superposition of multiple convolutional layers can extract higher-level feature representations; the operation formula of the convolutional layer is as follows:

[0077]

[0078] In the formula, F l (x, y) is the output feature value at position (x, y) in the l-th layer, and W l is the convolution kernel of the l-th layer (with size k h *k w, such as 5×5), F l-1 is the input feature map of the previous layer, b l is the bias term, and f(·) is the activation function;

[0079] (3) All 5 pooling layers in the CNN model adopt the max-pooling method, which selects the maximum value from a local area of the input feature map as the output to reduce the spatial dimension of the feature map, the number of parameters, and the computational complexity; the calculation formula of the max-pooling method is as follows:

[0080]

[0081] In the formula, P l (x, y) is the value of the output feature map of the max-pooling layer at position (x, y), k h *k w is the pooling window size (such as 4×4), and s is the sliding step;

[0082] (4) The fully connected layer is a multi-layer perceptron in which neurons between layers are all connected, located at the end of the model, used to integrate the features extracted by the convolutional layer and the pooling layer, and output the final prediction result.

[0083] Furthermore, the S4 specifically includes:

[0084] (1) Standardize the high-dimensional fully connected layer features obtained in S3 to eliminate the dimensional difference, and its processing formula is:

[0085]

[0086] Among them, μ is the feature mean vector, σ is the standard deviation vector, and the dimensional difference between features is eliminated:

[0087] (2) Based on the standardized data matrix Z, calculate the covariance matrix C to reveal the linear correlation between features; use singular value decomposition (SVD) to perform eigen-decomposition on C to obtain the eigenvalues λ i arranged in descending order of variance contribution rate and the corresponding orthogonal eigenvectors v i ;

[0088] (3) According to the cumulative variance contribution rate criterion, select the first 30 principal components so that their cumulative contribution rate is not less than 95%; construct the projection matrix W = [v 1 , v 2 , …, v 30 , and project the standardized features into the principal component space:

[0089] Y = ZW

[0090] In the formula, Y is the feature matrix after dimensionality reduction;

[0091] (4) Output the 30-dimensional feature matrix Y as the input of the SVM classifier, reducing the training time of the SVM by more than 90%.

[0092] Further, the S5 specifically includes:

[0093] (1) Input the 30-dimensional feature matrix Y ∈ R after PCA dimensionality reduction, where each row corresponds to a sample and each column corresponds to a principal component feature; N is the total number of samples, representing the number of all centrifugal pump status data; label data: L ∈ {1, 2, 3, 4}, and each label corresponds to a failure type (1 - normal, 2 - cavitation, 3 - impeller wear, 4 - bearing damage); N*30

[0094] (2) Perform parameter initialization, setting the GWO population size, the number of iterations, and the SVM parameter search range; where the penalty factor C ∈ [0, 100] is set to balance the classification error and the model complexity; the radial basis kernel function parameter σ ∈ [0, 100] is set to control the distribution of data mapped to the high-dimensional space; the GWO algorithm parameters are set, the population size N = 20, the maximum number of iterations T = 30, and the convergence factor a linearly decreases from 2 to 0 to control the search range to shift from global exploration to local exploitation;

[0095] (3) Randomly divide the training set samples into 5 parts. For the current gray wolf parameters (C, σ), train the SVM with 4 parts in turn and validate with 1 part, for a total of 5 cycles; record the accuracy of each validation, and take the average of the 5 times as the fitness value of this parameter combination:

[0096]

[0097] In the formula, Accuracy k is the classification accuracy of the k-th validation set;

[0098] (4) In the gray wolf optimization algorithm, the wolf pack is divided into four social ranks: α, β, δ, ω; sorted by the fitness value, the top 3 are α (optimal), β (sub-optimal), δ (third), and their parameter combinations are the current optimal solutions; each wolf adjusts its own parameters according to the positions of α, β, and δ:

[0099]

[0100] In the formula, X α , X β , X δ are the parameter combinations of the top three optimal wolves;

[0101] (5) When the maximum number of iterations is reached (30 times), terminate the algorithm and output the global optimal parameters C * and σ * ;

[0102] (6) The optimal parameter combination obtained by GWO is used to configure the SVM model, and the radial basis kernel (RBF) is selected to achieve nonlinear classification; based on the one-versus-rest (OvR) strategy, independent binary-class SVMs are trained for each fault category (normal, cavitation, impeller wear, bearing damage); the 30-dimensional features of the test set samples are input, and the decision values are calculated by 4 SVM models respectively, and the category with the largest decision value is selected as the prediction result; the confidence score is generated according to the normalized distance from the sample to the classification hyperplane, and the closer the value is to 1, the more reliable the classification result is.

[0103] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0104] The centrifugal pump fault diagnosis method, device and electronic equipment based on continuous wavelet transform and deep learning combine the units and algorithm steps of each example described in the embodiments disclosed in this article, and can be implemented by electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0105] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution.

[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0107] It should be noted that all directional indications (such as up, down, left, right, front, back, horizontal, vertical, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement conditions between components in a certain specific posture (as shown in the drawings). If this specific posture changes, the directional indications will also change accordingly. The "connection" can be a direct connection or an indirect connection, and the "setting", "set on", and "set in" can be directly set or indirectly set.

[0108] The above embodiments are illustrative of the present invention and not restrictive thereof. It can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The protection scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning, characterized in that: The following steps are involved: S1: Collect centrifugal multi-dimensional vibration data through a three-axis acceleration sensor, construct a data set through sliding window segmentation and data enhancement, and normalize the data set; S2: Perform continuous wavelet transform on the original vibration signal to obtain the time-frequency characteristics of the signal; randomly divide the training set and the test set into a ratio of 4:1 for model training and effect verification; S3: Use CNN to extract useful information for fault diagnosis from time-frequency graphs; use data regularization to reduce the differences between data and improve the generalization ability of the model; S4: PCA is used to extract the main components of the data to reduce the dimension of the data, reduce the computational complexity, and improve the performance of the model; S5: Use the training set feature data to train the SVM so that it can identify different types of faults; use GWO to optimize parameters and find the best combination of C and σ to improve the classification accuracy of the SVM.

2. According to claim 1, a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning is characterized in that: In step S1, the specific process includes the following steps: (1) Normalizing the original vibration signal of the centrifugal pump collected in S1; (2) Overlap sampling is performed on the normalized vibration data obtained in step (1) to increase the number of training samples, with an overlap rate of 0.5; then, sample labels are generated based on the state of the centrifugal pump to which each data sample belongs, and the vibration data used includes three fault states and one normal state.

3. According to claim 1, a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning is characterized in that: In step S2, the specific process includes the following steps: (1) Perform continuous wavelet transform on the data in the sample set obtained in step S1. The formula of continuous wavelet transform is as follows: Where W f (a, τ) is the output result of wavelet transform, which is two-dimensional data; f(t) is the input signal, which is the complex conjugate function of the wavelet basis function, a is the scale factor, τ is the translation factor, and this patent uses Morlet wavelet as the basis function; (2) The result obtained after continuous wavelet transform contains amplitude information amp and frequency information f. The time axis t is generated according to the set sampling frequency and sample length, and the time-frequency diagram corresponding to each sample is drawn based on amp, f and tt; (3) Save the time-frequency graph obtained in step (2) in PNG format, and convert each time-frequency graph into a 280×280 RGB format image to meet the training requirements of the convolutional neural network constructed later, where 280×280 represents the pixel size of the time-frequency graph; finally, divide these images into a training set and a test set in a ratio of 4:

1.

4. According to claim 1, a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning is characterized in that: In step S3, the specific process includes the following steps: (1) A convolutional neural network model was built based on Python language and PyTorch platform, including convolutional layers, pooling layers and fully connected layers. The specific network structure includes 5 convolutional layers, 5 pooling layers and 1 fully connected layer. The convolutional layers are divided into shallow convolution and deep convolution. The pooling layers all use the maximum pooling method. The fully connected layer flattens the feature vector into a one-dimensional vector. The fully connected layer uses the Softmax function activation to output the probability of each category. (2) The convolution layer performs convolution operations on the input feature map to extract the input features. The superposition of multiple convolution layers can extract higher-level feature representations. The operation formula of the convolution layer is as follows: Where F l (x, y) is the output feature value of the lth layer at position (x, y), W l is the convolution kernel of layer l (size k h *k w , such as 5×5), F l-1 is the input feature map of the previous layer, b l is the bias term, f(·) is the activation function; (3) The five pooling layers in the CNN model all use the maximum pooling method, which selects the maximum value from a local area of ​​the input feature map as the output, which is used to reduce the spatial dimension of the feature map, reduce the number of parameters and computational complexity; the calculation formula of the maximum pooling method is as follows: Where P l (x, y) is the value of the maximum pooling layer output feature map at position (x, y), k h *k w is the pooling window size (e.g. 4×4), and s is the sliding step size; (4) The fully connected layer is a multilayer perceptron in which all neurons between layers are connected. It is located at the end of the model and is used to integrate the features extracted by the convolutional layer and the pooling layer and output the final prediction results.

5. According to claim 1, a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning is characterized in that: In step S4, the specific process includes the following steps: (1) The high-dimensional fully connected layer features obtained in step 4 are standardized to eliminate the dimension difference. The processing formula is: Among them, μ is the feature mean vector, σ is the standard deviation vector, which eliminates the dimensional differences between features; (2) Based on the standardized data matrix Z, the covariance matrix C is calculated to reveal the linear correlation between features; singular value decomposition (SVD) is used to decompose C to obtain the eigenvalues ​​λ arranged in descending order of variance contribution rate i and the corresponding orthogonal eigenvector v i ; (3) According to the cumulative variance contribution rate criterion, the first 30 principal components are selected so that their cumulative contribution rate is not less than 95%; Construct a projection matrix to project the standardized features into the principal component space; (4) Output a 30-dimensional feature matrix as the input of the SVM classifier, reducing the SVM training time by more than 90%.

6. According to claim 1, a centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning is characterized in that: In step S5, the specific process includes the following steps: (1) Input the 30-dimensional feature matrix Y∈R after PCA dimensionality reduction N*30 , where each row corresponds to a sample and each column corresponds to a principal component feature; N is the total number of samples, indicating the number of all centrifugal pump status data; label data: L∈{1,2,3,4}, each label corresponds to a fault type (1-normal, 2-cavitation, 3-impeller wear, 4-bearing damage); (2) Initialize parameters, set the GWO population size, number of iterations, and SVM parameter search range; The penalty factor C∈[0,100] is set to balance the classification error and model complexity; the radial basis kernel function parameter σ∈[0,100] is set to control the distribution of data mapped to high-dimensional space; the GWO algorithm parameters are set, the population size is 20, the maximum number of iterations is 30, and the convergence factor a decreases linearly from 2 to 0, controlling the search range from global exploration to local development; (3) The training set samples are randomly divided into 5 parts. For the current gray wolf parameters (C, σ), 4 parts are used to train the SVM and 1 part is used for verification, for a total of 5 cycles; the accuracy of each verification is recorded, and the average of 5 times is taken as the fitness value of this parameter combination: Where Accuracy k is the classification accuracy of the k-th validation set, and Fitness refers to the fitness; (4) In the gray wolf optimization algorithm, the wolf pack is divided into four social levels: α, β, δ, and ω; sorted by fitness value, the top three are α (optimal), β (second-best), and δ (third), and their parameter combination is the current optimal solution; each wolf adjusts its own parameters according to the position of α, β, and δ: Where, X α , X β , X δ It is the parameter combination of the top three best wolves; (5) When the maximum number of iterations (30) is reached, the algorithm is terminated and the global optimal parameter C is output. * and σ * ; (6) The optimal parameter combination obtained by GWO optimization is used to configure the SVM model, and the radial basis kernel (RBF) is selected to realize nonlinear classification. Based on the one-vs-many (OvR) strategy, an independent binary classification SVM is trained for each fault category (normal, cavitation, impeller wear, and bearing damage). The 30-dimensional features of the test set samples are input, and the decision values ​​are calculated respectively by four SVM models. The category with the largest decision value is selected as the prediction result. The confidence score is generated according to the normalized distance from the sample to the classification hyperplane. The closer the value is to 1, the more reliable the classification result is.

7. A centrifugal pump fault diagnosis device based on continuous wavelet transform and deep learning, characterized in that: include: The data processing module is used to collect the original vibration data of various faults of the centrifugal pump under known working conditions and perform normalization processing on them; Overlap sampling is performed on the normalized data to obtain data samples of fixed length, labels are created for each sample based on the fault type, and the labeled samples are divided into a training sample set and a test sample set; The data set preparation module is used to perform continuous wavelet transform on the training samples and test samples obtained in step 1 to obtain the time-frequency diagram of each sample for the training of the convolutional neural network constructed subsequently; Model pre-training module, used to build convolutional neural network models based on Python and PyTorch deep learning framework; The diagnosis module is executed, PCA is used to reduce the dimension of the data, and the feature data of the training set is used to train the SVM so that it can identify different types of faults. GWO is used to optimize the parameters to improve the classification accuracy of the SVM. The fault diagnosis data in the form of a time-frequency diagram of the centrifugal pump is input to obtain the fault diagnosis results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning as described in any one of claims 1 to 5 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the centrifugal pump fault diagnosis method based on continuous wavelet transform and deep learning as described in any one of claims 1 to 5 is implemented.

Images