A centrifugal pump fault diagnosis system based on artificial intelligence

Abstract

The present invention discloses a centrifugal pump fault diagnosis system based on artificial intelligence, which relates to the field of centrifugal pump fault diagnosis technology, including a data acquisition module, a data preprocessing module, a feature extraction and selection module, a deep learning diagnosis model, a multi-fault type identification module and a self-learning and model update module; through a multi-level cross feature extraction and optimization selection algorithm, features can be extracted more comprehensively from the preprocessed data; by collecting historical operation data and fault label data, a deep learning model is used to construct and train a fault diagnosis model; a multi-dimensional hybrid power optimization algorithm performs multi-level complex calculations and unique adjustment mechanism updates on model parameters, ensuring efficient optimization and stable convergence of the model. The optimization algorithm introduces historical gradient information, nonlinear adjustment and hybrid adjustment, making the model parameter update more robust and efficient, and improving the model diagnosis accuracy and convergence speed.

Description

Technical Field

[0001] The invention belongs to the technical field of centrifugal pump fault diagnosis, and in particular relates to a centrifugal pump fault diagnosis system based on artificial intelligence. Background Art

[0002] In the industrial field, centrifugal pumps are important mechanical equipment and are widely used in chemical, petroleum, metallurgy, and electric power industries. The failure of a centrifugal pump will directly affect the stability and safety of the production process. Therefore, real-time monitoring and diagnosis of centrifugal pump failures are of great significance. Traditional centrifugal pump fault diagnosis methods mainly rely on manual experience and simple statistical analysis. These methods have the following limitations: manual experience depends on the experience and knowledge level of the operator, and the judgments of different operators may vary greatly, making it difficult to ensure the consistency and reliability of diagnosis. With the complexity of industrial systems, it is difficult to fully and accurately diagnose centrifugal pump failures based solely on manual experience. Traditional statistical analysis methods such as time domain analysis or frequency domain analysis can extract and analyze some fault features, but they are unable to cope with high-dimensional multivariate data. The above methods can usually only process a single or a small number of features and cannot fully reflect the operating status of the equipment. In addition, these methods have limited ability to handle data noise and inconsistency, and are easily interfered by external factors, resulting in inaccurate fault diagnosis results. Traditional fault diagnosis methods often cannot achieve real-time monitoring and diagnosis, and it is difficult to detect and diagnose faults in real time and accurately during the operation of centrifugal pumps. Summary of the invention

[0003] The purpose of the present invention is to provide a centrifugal pump fault diagnosis system based on artificial intelligence to solve the problems raised by the above background technology.

[0004] An artificial intelligence-based centrifugal pump fault diagnosis system of the present invention comprises a centrifugal pump and fault diagnosis, wherein the fault diagnosis comprises a data acquisition module, a data preprocessing module, a feature extraction and selection module, a deep learning diagnosis model, a multi-fault type identification module and a self-learning and model updating module;

[0005] The data acquisition module is used to collect the operating data of the centrifugal pump and send the operating data to the data preprocessing module; the data preprocessing module preprocesses the original operating data to obtain preprocessed data, and sends the preprocessed data to the feature extraction and selection module; the feature extraction and selection module extracts features from the preprocessed data, screens the extracted features, obtains key feature data, and sends the key feature data to the deep learning diagnosis model; the deep learning diagnosis model is used to optimize the deep learning fault diagnosis model to obtain preliminary diagnosis results, and send the preliminary diagnosis results to the multi-fault type identification module; the multi-fault type identification module is used to accurately identify specific fault types and corresponding fault severity to obtain final diagnosis results; the self-learning and model update module updates the model parameters of the deep learning fault diagnosis model to adapt to new operating conditions and fault types.

[0006] Furthermore, the fault diagnosis steps are as follows:

[0007] S1. Collect the operation data of the centrifugal pump through sensors to obtain the original operation data, and pre-process the original operation data by cleaning, denoising, smoothing and standardization to obtain the pre-processed data; use a multi-level cross-feature extraction and optimization selection algorithm to extract and screen the pre-processed data to obtain key features; S2. Train the deep learning fault diagnosis model through the historical operation data of the centrifugal pump, and update the model parameters of the deep learning fault diagnosis model to obtain an optimized deep learning fault diagnosis model. Perform fault diagnosis on the key feature data through the optimized deep learning fault diagnosis model to obtain preliminary diagnosis results. Based on the preliminary diagnosis results, refine the fault type to obtain the final diagnosis result.

[0008] Furthermore, the multi-level cross feature extraction and optimization selection algorithm extracts key features from the pre-processed centrifugal pump operation data through multi-level cross processing, and screens the key features to obtain the most representative feature data; the specific implementation process is as follows:

[0009] The preprocessed data matrix is ​​subjected to statistical analysis and feature engineering technology to perform preliminary feature extraction to obtain a preliminary feature data set. The first-layer feature extraction is performed on the preliminary feature data set using a nonlinear transformation method. The nonlinear transformation extracts the nonlinear features of the data through complex mathematical transformations. The formula is as follows:

[0010]

[0011] in, Indicates The sample nonlinear transformation eigenvalues; the sample is a preliminary feature data set obtained by performing preliminary feature extraction on the preprocessed centrifugal pump operation data; Indicates the number of samples; Indicates The sample The first eigenvalues; It represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction; Indicates The standard deviation of the preliminary features reflects the degree of dispersion of the data; Indicates The mean of the preliminary features, indicating the central tendency of the data; It means that the influence of high variance features is adjusted and suppressed by negative exponential transformation of variance, the contribution of different features is balanced to reduce noise interference, and the nonlinear expression ability of feature extraction process is enhanced; Indicates The variance of the initial features, indicating the volatility of the data; is a small positive number to prevent the denominator from being zero;

[0012] The weighted polynomial method is used to extract the second-level features of the preliminary feature data set. The weighted polynomial feature extraction aims to extract the polynomial feature formula of the data through polynomial transformation and weighted processing as follows:

[0013]

[0014] in, Indicates The sample Weighted polynomial eigenvalues; weighted polynomial feature extraction captures the polynomial relationship in the data through polynomial transformation and enhances the expressiveness of the features; Represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction and is used to adjust the importance of each feature in the weighted polynomial feature extraction process. Indicates The mean of the preliminary features indicates the central tendency of the data.

[0015] The formula for cross-processing the features extracted from the first and second layers by the cross-feature processing method is as follows:

[0016]

[0017] in, Indicates The sample Cross eigenvalues; To adjust parameters and control the balance of features at different levels; the technology of cross-feature processing is to enhance the comprehensive expression ability of features by combining features at different levels.

[0018] Furthermore, feature importance measurement screening is performed based on cross-features, and the following formula is used to measure the importance of feature selection:

[0019]

[0020] in, Indicates The importance of each feature; screening key features is to calculate the importance of features and select the most representative features; is a small constant that prevents the denominator from being zero; is a tuning parameter used to control the effect of eigenvalues ​​on importance; is the threshold of the feature, which is used to control the strictness of feature screening;

[0021] According to the importance of features , sort the features and select the first The most important features are taken as key features; the key feature data output constitutes the final key feature data matrix, providing high-quality data input for the subsequent deep learning diagnostic model.

[0022] Furthermore, historical operation data and fault label data of the centrifugal pump are collected, an existing deep learning model is selected to construct a deep learning fault diagnosis model, and the deep learning fault diagnosis model is trained using the historical operation data and fault label data, the training goal is to minimize loss functions such as cross entropy loss and mean square error loss; the deep learning fault diagnosis model is trained;

[0023] In order to obtain an optimized deep learning fault diagnosis model, a multi-dimensional hybrid power optimization algorithm is used to update the model parameters; the multi-dimensional hybrid power optimization algorithm updates and optimizes the model parameters through multi-level complex calculations and a unique adjustment mechanism to ensure efficient optimization and stable convergence of the model; the specific implementation process is as follows,

[0024] Initialize model parameters , set the learning rate , Momentum Factor , nonlinear adjustment factor , mixed adjustment factor and stability constant Model parameters Random initialization to break the symmetry; learning rate Controls the step size of parameter updates, affecting training speed and stability; momentum factor Controlling the momentum gradient The smoothness of the curve helps accelerate convergence and reduce oscillation; nonlinear adjustment factor Used to control the degree of nonlinear regulation, mixed regulation factor Controls the weight of the mixed regulation, stability constant Prevent division by zero operations to ensure calculation stability;

[0025] In each round of model parameter update iteration, in order to know the direction and magnitude of parameter update, the loss function is calculated Relative to model parameters Gradient , the gradient calculation formula is as follows,

[0026]

[0027] in, is the loss function Relative to model parameters The gradient of The gradient value during the round iteration; the larger the gradient, the greater the adjustment of the parameters required;

[0028] Calculate the momentum gradient ; By introducing historical gradient information, the fluctuation of the current gradient is smoothed, thereby accelerating convergence; Momentum gradient The calculation formula is as follows,

[0029]

[0030] Here is the momentum gradient of the previous iteration , is the current gradient.

[0031] Furthermore, the introduction of the nonlinear adjustment factor performs nonlinear adjustment on the momentum gradient and calculates the nonlinear adjustment gradient ; Nonlinear adjustment is to adjust the amplitude and direction of the gradient through nonlinear transformation, making parameter update more flexible; the calculation formula of nonlinear adjustment gradient is as follows,

[0032]

[0033] in, is an exponential function, is the hyperbolic tangent function, is a nonlinear adjustment factor; Partly through exponential functions and nonlinear adjustment factors Momentum gradient Perform nonlinear adjustment. Part of it is the hyperbolic tangent function of the current gradient Perform non-linear adjustment.

[0034] Furthermore, the nonlinear adjustment gradient is followed by a mixed adjustment to calculate the mixed gradient ; Mixed adjustment is to integrate the information of momentum gradient and current gradient, making the gradient update more robust; Mixed gradient The calculation formula is as follows,

[0035]

[0036] in, is a logarithmic function, is a mixed adjustment factor; Part of it is the nonlinear adjustment gradient Logarithmic transformation and weighting are performed. Part of it is the current gradient Perform momentum correction and weighting;

[0037] In order to correct the mixed gradient, calculate the corrected gradient ; The correction is to make the gradient smoother and more robust by combining momentum and nonlinear adjustment; the calculation formula of the corrected gradient is as follows,

[0038]

[0039] in, is a stability constant to avoid the denominator being zero; the mixed gradient Perform square mean correction to make gradient update smoother and more robust; square mean estimation The calculation formula is as follows,

[0040]

[0041] Through the momentum factor Smooth the square of the current gradient and get the square mean .

[0042] Furthermore, according to the corrected gradient Update model parameters ; The purpose of parameter updating is to adjust the parameters so that the loss function Minimize; the calculation formula for parameter update is as follows,

[0043]

[0044] The above formula is obtained by learning rate and the corrected gradient For current parameters Update and get the parameters for the next iteration ;

[0045] When the preset threshold or the preset number of iterations is reached, the iteration ends;

[0046] At the same time, during the training process of the deep learning fault diagnosis model, the cross-validation method is used to verify the model, and finally a trained and verified optimized deep learning fault diagnosis model is obtained.

[0047] The beneficial effects of the present invention are:

[0048] Through multi-level cross feature extraction and optimization selection algorithms, features can be extracted more comprehensively from pre-processed data and scientifically screened; the algorithm combines multi-dimensional processing such as nonlinear transformation, weighted polynomials and cross processing to ensure the representativeness and importance of features, providing a high-quality data source for subsequent deep learning diagnostic models;

[0049] By collecting historical operation data and fault label data, the deep learning model is used to build and train the fault diagnosis model; the multi-dimensional hybrid power optimization algorithm performs multi-level complex calculations and a unique update mechanism on the model parameters, ensuring efficient optimization and stable convergence of the model; the optimization algorithm introduces historical gradient information, nonlinear adjustment and hybrid adjustment to make the model parameter update more robust and efficient, significantly improving the diagnostic accuracy and convergence speed of the model; thereby improving the overall accuracy of the artificial intelligence-based centrifugal pump fault diagnosis system in the diagnosis process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a structural diagram of a centrifugal pump fault diagnosis system based on artificial intelligence according to the present invention;

[0051] Figure 2 This is a diagnostic step diagram of a centrifugal pump fault diagnosis system based on artificial intelligence described in the present invention. DETAILED DESCRIPTION

[0052] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0054] The specific scheme of the centrifugal pump fault diagnosis system based on artificial intelligence provided by the present invention is described in detail below with reference to the accompanying drawings.

[0055] See attached Figure 1-Figure 2 , which shows an artificial intelligence-based centrifugal pump fault diagnosis system provided by an embodiment of the present invention, including a centrifugal pump and fault diagnosis, wherein the fault diagnosis includes a data acquisition module, a data preprocessing module, a feature extraction and selection module, a deep learning diagnosis model, a multi-fault type identification module and a self-learning and model updating module;

[0056] The data acquisition module is used to collect the operation data of the centrifugal pump, obtain the original operation data, and send the original operation data to the data preprocessing module; the data preprocessing module preprocesses the original operation data to obtain preprocessed data, and sends the preprocessed data to the feature extraction and selection module; the feature extraction and selection module extracts features from the preprocessed data, screens the extracted features, obtains key feature data, and sends the key feature data to the deep learning diagnosis model; the deep learning diagnosis model trains the deep learning fault diagnosis model through the historical operation data of the centrifugal pump, obtains the optimized deep learning fault diagnosis model by receiving the optimized model parameters, performs fault diagnosis based on the key feature data through the optimized deep learning fault diagnosis model, obtains a preliminary diagnosis result, and sends the preliminary diagnosis result to the multi-fault type identification module; the multi-fault type identification module refines the fault type according to the preliminary diagnosis result, accurately identifies the specific fault type and the corresponding fault severity, and obtains the final diagnosis result; the self-learning and model updating module updates the model parameters of the deep learning fault diagnosis model to adapt to the new operating conditions and fault types, and obtains the optimized model parameters.

[0057] The troubleshooting steps are as follows:

[0058] S1. Collect the operation data of the centrifugal pump through the sensor to obtain the original operation data, perform preprocessing of cleaning, denoising, smoothing and standardization on the original operation data to obtain the preprocessed data; perform feature extraction and screening on the preprocessed data using a multi-level cross feature extraction and optimization selection algorithm to obtain key features;

[0059] S2. Train the deep learning fault diagnosis model through the historical operation data of the centrifugal pump, and update the model parameters of the deep learning fault diagnosis model to obtain an optimized deep learning fault diagnosis model. Perform fault diagnosis on the key feature data through the optimized deep learning fault diagnosis model to obtain preliminary diagnosis results. Based on the preliminary diagnosis results, refine the fault type to obtain the final diagnosis result.

[0060] The multi-level cross feature extraction and optimization selection algorithm extracts key features from the pre-processed centrifugal pump operation data through multi-level cross processing, and screens the key features to obtain the most representative feature data; the specific implementation process is as follows:

[0061] The preprocessed data matrix is ​​subjected to statistical analysis and feature engineering technology to perform preliminary feature extraction to obtain a preliminary feature data set. The nonlinear transformation method is used to extract the first layer of features from the preliminary feature data set. The nonlinear transformation extracts the nonlinear features of the data through complex mathematical transformations. The formula is as follows:

[0062]

[0063] in, Indicates The sample nonlinear transformation eigenvalues; the sample is a preliminary feature data set obtained by performing preliminary feature extraction on the preprocessed centrifugal pump operation data; Indicates the number of samples; Indicates The sample The first eigenvalues; It represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction; Indicates The standard deviation of the preliminary features reflects the degree of dispersion of the data; Indicates The mean of the preliminary features, indicating the central tendency of the data; It means that the influence of high variance features is adjusted and suppressed by negative exponential transformation of variance, the contribution of different features is balanced to reduce noise interference, and the nonlinear expression ability of feature extraction process is enhanced; Indicates The variance of the initial features, indicating the volatility of the data; is a small positive number to prevent the denominator from being zero;

[0064] The weighted polynomial method is used to extract the second-level features of the preliminary feature data set. The weighted polynomial feature extraction aims to extract the polynomial feature formula of the data through polynomial transformation and weighted processing as follows:

[0065]

[0066] in, Indicates The sample Weighted polynomial eigenvalues; weighted polynomial feature extraction captures the polynomial relationship in the data through polynomial transformation and enhances the expressiveness of the features; Represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction and is used to adjust the importance of each feature in the weighted polynomial feature extraction process. Indicates The mean of the preliminary features indicates the central tendency of the data.

[0067] The formula for cross-processing the features extracted from the first and second layers by the cross-feature processing method is as follows:

[0068]

[0069] in, Indicates The sample Cross eigenvalues; To adjust parameters and control the balance of features at different levels; the technology of cross-feature processing is to enhance the comprehensive expression ability of features by combining features at different levels.

[0070] Based on the cross-features, feature importance measurement is screened and the following formula is used to measure the importance of feature selection:

[0071]

[0072] in, Indicates The importance of each feature; screening key features is to calculate the importance of features and select the most representative features; is a small constant that prevents the denominator from being zero; is a tuning parameter used to control the effect of eigenvalues ​​on importance; is the threshold of the feature, which is used to control the strictness of feature screening;

[0073] According to the importance of features , sort the features and select the first The most important features are taken as key features; the key feature data output constitutes the final key feature data matrix, providing high-quality data input for the subsequent deep learning diagnostic model.

[0074] Collect historical operation data and fault label data of centrifugal pumps, select existing deep learning models (such as long short-term memory network models, convolutional neural networks, hybrid models, recurrent neural networks, and models based on attention mechanisms) to build deep learning fault diagnosis models, and use historical operation data and fault label data to train the deep learning fault diagnosis models. The training goal is to minimize loss functions such as cross entropy loss and mean square error loss; train the deep learning fault diagnosis model;

[0075] In order to obtain an optimized deep learning fault diagnosis model, a multi-dimensional hybrid power optimization algorithm is used to update the model parameters; the multi-dimensional hybrid power optimization algorithm updates and optimizes the model parameters through multi-level complex calculations and a unique adjustment mechanism to ensure efficient optimization and stable convergence of the model; the specific implementation process is as follows,

[0076] Initialize model parameters , set the learning rate , Momentum Factor , nonlinear adjustment factor , mixed adjustment factor and stability constant Model parameters Random initialization to break the symmetry; learning rate Controls the step size of parameter updates, affecting training speed and stability; momentum factor Controlling the momentum gradient The smoothness of the curve helps accelerate convergence and reduce oscillation; nonlinear adjustment factor Used to control the degree of nonlinear regulation, mixed regulation factor Controls the weight of the mixed regulation, stability constant Prevent division by zero operations to ensure calculation stability;

[0077] In each round of model parameter update iteration, in order to know the direction and magnitude of parameter update, the loss function is calculated Relative to model parameters Gradient , the gradient calculation formula is as follows,

[0078]

[0079] in, is the loss function Relative to model parameters The gradient of The gradient value during the round iteration; the larger the gradient, the greater the adjustment of the parameters required;

[0080] Calculate the momentum gradient ; By introducing historical gradient information, the fluctuation of the current gradient is smoothed, thereby accelerating convergence; Momentum gradient The calculation formula is as follows,

[0081]

[0082] Here is the momentum gradient of the previous iteration , is the current gradient.

[0083] The nonlinear adjustment factor is introduced to perform nonlinear adjustment on the momentum gradient and calculate the nonlinear adjustment gradient ; Nonlinear adjustment is to adjust the amplitude and direction of the gradient through nonlinear transformation, making parameter update more flexible; the calculation formula of nonlinear adjustment gradient is as follows,

[0084]

[0085] in, is an exponential function, is the hyperbolic tangent function, is a nonlinear adjustment factor; Partly through exponential functions and nonlinear adjustment factors Momentum gradient Perform nonlinear adjustment. Part of it is the hyperbolic tangent function of the current gradient Perform non-linear adjustment.

[0086] The nonlinear adjustment gradient is followed by mixed adjustment to calculate the mixed gradient ; Mixed adjustment is to integrate the information of momentum gradient and current gradient, making the gradient update more robust; Mixed gradient The calculation formula is as follows,

[0087]

[0088] in, is a logarithmic function, is a mixed adjustment factor; Part of it is the nonlinear adjustment gradient Logarithmic transformation and weighting are performed. Part of it is the current gradient Perform momentum correction and weighting;

[0089] In order to correct the mixed gradient, calculate the corrected gradient ; The correction is to make the gradient smoother and more robust by combining momentum and nonlinear adjustment; the calculation formula of the corrected gradient is as follows,

[0090]

[0091] in, is a stability constant to avoid the denominator being zero; the mixed gradient Perform square mean correction to make gradient update smoother and more robust; square mean estimation The calculation formula is as follows,

[0092]

[0093] Through the momentum factor Smooth the square of the current gradient and get the square mean .

[0094] According to the corrected gradient Update model parameters ; The purpose of parameter updating is to adjust the parameters so that the loss function Minimize; the calculation formula for parameter update is as follows,

[0095]

[0096] The above formula is obtained by learning rate and the corrected gradient For current parameters Update and get the parameters for the next iteration ;

[0097] When the preset threshold or the preset number of iterations is reached, the iteration ends;

[0098] Through the above process, a deep learning fault diagnosis model was built and optimized to ensure that the model can accurately and real-time identify centrifugal pump faults and dynamically update and optimize according to new data.

[0099] At the same time, during the training process of the deep learning fault diagnosis model, the cross-validation method is used to verify the model, and finally a trained and verified optimized deep learning fault diagnosis model is obtained.

[0100] Fault diagnosis is performed on key feature data through an optimized deep learning fault diagnosis model to obtain preliminary diagnosis results, which represent the fault probability distribution corresponding to each sample. Based on the preliminary diagnosis results, the existing decision tree is used to refine the fault type to obtain the final diagnosis result, thereby realizing artificial intelligence-based centrifugal pump fault diagnosis.

[0101] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0102] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A centrifugal pump fault diagnosis system based on artificial intelligence, comprising a centrifugal pump and fault diagnosis, characterized in that: The fault diagnosis includes a data acquisition module, a data preprocessing module, a feature extraction and selection module, a deep learning diagnosis model, a multi-fault type identification module and a self-learning and model updating module; The data acquisition module is used to collect the operation data of the centrifugal pump and send the operation data to the data preprocessing module; The data preprocessing module preprocesses the original operating data to obtain preprocessed data, and sends the preprocessed data to the feature extraction and selection module; the feature extraction and selection module extracts features from the preprocessed data, uses a multi-level cross feature extraction and optimization selection algorithm to extract and screen the preprocessed data, and screens the extracted features to obtain key feature data, and sends the key feature data to the deep learning diagnosis model; the deep learning diagnosis model is used to optimize the deep learning fault diagnosis model to obtain preliminary diagnosis results, and send the preliminary diagnosis results to the multi-fault type identification module; the multi-fault type identification module is used to accurately identify specific fault types and corresponding fault severity to obtain final diagnosis results; the self-learning and model update module updates the model parameters of the deep learning fault diagnosis model to adapt to new operating conditions and fault types; The multi-level cross feature extraction and optimization selection algorithm extracts key features from the pre-processed centrifugal pump operation data through multi-level cross processing, and screens the key features to obtain the most representative feature data; the specific implementation process is as follows: The preprocessed data matrix is ​​subjected to statistical analysis and feature engineering technology to perform preliminary feature extraction to obtain a preliminary feature data set, and the first-layer feature extraction is performed on the preliminary feature data set using a nonlinear transformation method; Nonlinear transformation extracts the nonlinear characteristics of data through complex mathematical transformation. The formula is as follows: ； in, Indicates The sample nonlinear transformation eigenvalues; the sample is a preliminary feature data set obtained by performing preliminary feature extraction on the preprocessed centrifugal pump operation data; Indicates the number of samples; Indicates The sample The first eigenvalues; It represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction; Indicates The standard deviation of the preliminary features reflects the degree of dispersion of the data; Indicates The mean of the preliminary features, indicating the central tendency of the data; It means that the influence of high variance features is adjusted and suppressed by negative exponential transformation of variance, the contribution of different features is balanced to reduce noise interference, and the nonlinear expression ability of feature extraction process is enhanced; Indicates The variance of the initial features, indicating the volatility of the data; is a small positive number to prevent the denominator from being zero; The weighted polynomial method is used to extract the second-level features of the preliminary feature data set. The weighted polynomial feature extraction aims to extract the polynomial features of the data through polynomial transformation and weighted processing. The formula is as follows: ； in, Indicates The sample Weighted polynomial eigenvalues; weighted polynomial feature extraction captures the polynomial relationship in the data through polynomial transformation and enhances the expressiveness of the features; Represents the weighting coefficient, which reflects the contribution of different preliminary features to feature extraction and is used to adjust the importance of each feature in the weighted polynomial feature extraction process. Indicates The mean of the preliminary features indicates the central tendency of the data. The cross-feature processing method is used to cross-process the features extracted from the first layer and the second layer. The formula is as follows: ； in, Indicates The sample Cross eigenvalues; To adjust parameters and control the balance of features at different levels; the technology of cross-feature processing is to enhance the comprehensive expression ability of features by combining features at different levels.

2. According to claim 1, a centrifugal pump fault diagnosis system based on artificial intelligence is characterized in that: The troubleshooting steps are as follows: S1. Collecting the operation data of the centrifugal pump through the sensor to obtain the original operation data, and pre-processing the original operation data by cleaning, denoising, smoothing and standardizing to obtain the pre-processed data; The preprocessed data is subjected to feature extraction and screening using a multi-level cross feature extraction and optimization selection algorithm to obtain key features; S2. Train the deep learning fault diagnosis model through the historical operation data of the centrifugal pump, and update the model parameters of the deep learning fault diagnosis model to obtain an optimized deep learning fault diagnosis model. Perform fault diagnosis on the key feature data through the optimized deep learning fault diagnosis model to obtain preliminary diagnosis results. Based on the preliminary diagnosis results, refine the fault type to obtain the final diagnosis result.

3. The centrifugal pump fault diagnosis system based on artificial intelligence according to claim 1 is characterized in that: Based on the cross-features, feature importance measurement is screened and the following formula is used to measure the importance of feature selection: ； in, Indicates The importance of each feature; screening key features is to calculate the importance of features and select the most representative features; is a small constant that prevents the denominator from being zero; is a tuning parameter used to control the effect of eigenvalues ​​on importance; is the threshold of the feature, which is used to control the strictness of feature screening; According to the importance of features , sort the features and select the first The most important features are taken as key features; the key feature data output constitutes the final key feature data matrix, providing high-quality data input for the subsequent deep learning diagnostic model.

4. The centrifugal pump fault diagnosis system based on artificial intelligence according to claim 2 is characterized in that: Collect historical operation data and fault label data of centrifugal pumps, select existing deep learning models to build deep learning fault diagnosis models, and use historical operation data and fault label data to train the deep learning fault diagnosis models. The training goal is to minimize the loss function, which includes cross entropy loss and mean square error loss; train the deep learning fault diagnosis model; In order to obtain an optimized deep learning fault diagnosis model, a multi-dimensional hybrid power optimization algorithm is used to update the model parameters; the specific implementation process is as follows: Initialize model parameters , set the learning rate , Momentum Factor , nonlinear adjustment factor , mixed adjustment factor and stability constant ; Model parameters Random initialization to break the symmetry; learning rate Controls the step size of parameter updates, affecting training speed and stability; momentum factor Controlling the momentum gradient The smoothness of , which helps accelerate convergence and reduce oscillation; Nonlinear adjustment factor Used to control the degree of nonlinear regulation, mixed regulation factor Controls the weight of the mixed regulation, stability constant Prevent division by zero operations to ensure calculation stability; In each round of model parameter update iteration, in order to obtain the direction and magnitude of parameter update, the loss function is calculated Relative to model parameters Gradient , the gradient calculation formula is as follows, ； in, is the loss function Relative to model parameters The gradient of The gradient value during the round iteration; the larger the gradient, the greater the adjustment of the parameters required; Calculate the momentum gradient ; By introducing historical gradient information, the fluctuation of the current gradient is smoothed, thereby accelerating convergence; Momentum gradient The calculation formula is as follows, ； Here is the momentum gradient of the previous iteration , is the current gradient.

5. The centrifugal pump fault diagnosis system based on artificial intelligence according to claim 4 is characterized in that: The introduction of the nonlinear adjustment factor has an impact on the momentum gradient Perform nonlinear adjustment and calculate nonlinear adjustment gradient ; The calculation formula of nonlinear adjustment gradient is as follows, ； in, is an exponential function, is the hyperbolic tangent function, is a nonlinear adjustment factor; Partly through exponential functions and nonlinear adjustment factors Momentum gradient Perform nonlinear adjustment. Part of it is the hyperbolic tangent function of the current gradient Perform non-linear adjustment.

6. The centrifugal pump fault diagnosis system based on artificial intelligence according to claim 5 is characterized in that: According to the momentum gradient after nonlinear adjustment Perform mixed adjustment to calculate the mixed gradient , mixed gradient The calculation formula is as follows, ； in, is a logarithmic function, is a mixed adjustment factor; Part of it is the nonlinear adjustment gradient Logarithmic transformation and weighting are performed. Part of it is the current gradient Perform momentum correction and weighting; To correct the mixed gradient, the corrected gradient needs to be calculated , the corrected gradient The calculation formula is as follows, ； in, is a stability constant to avoid the denominator being zero; the mixed gradient Perform square mean correction to make gradient update smoother and more robust; square mean The calculation formula is as follows, ； Through the momentum factor Smooth the square of the current gradient and get the square mean .

7. The centrifugal pump fault diagnosis system based on artificial intelligence according to claim 6 is characterized in that: According to the corrected gradient Update model parameters ; Make the loss function Minimize; the calculation formula for updating model parameters is as follows, ； The above formula is obtained by learning rate and the corrected gradient For current parameters Update and get the parameters for the next iteration ; When the preset threshold or the preset number of iterations is reached, the iteration ends; At the same time, during the training process of the deep learning fault diagnosis model, the cross-validation method is used to verify the model, and finally a trained and verified optimized deep learning fault diagnosis model is obtained.

Images