Centrifugal pump fault diagnosis method and device based on conditional generative adversarial network

Abstract

An embodiment of the present invention provides a centrifugal pump fault diagnosis method and device based on a conditional generative adversarial network, belonging to the technical field of equipment fault diagnosis. By acquiring the real-time vibration signal of the centrifugal pump bearing; preprocessing the real-time vibration signal to obtain the data to be diagnosed, and then inputting the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained by judging the data balance of the sample data to obtain unbalanced data, and using the trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; and then training a convolutional neural network using the virtual samples and the sample data. It effectively solves the problem of low recognition accuracy of centrifugal pump bearing faults caused by the unbalanced distribution of data in the existing methods. Different process schemes are adopted for balanced data and unbalanced data, greatly improving the speed of model training and also improving the accuracy of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault diagnosis, and particularly to a centrifugal pump fault diagnosis method based on a conditional generative adversarial network, a centrifugal pump fault diagnosis device based on a conditional generative adversarial network, a machine-readable storage medium, and a processor. Background Art

[0002] The centrifugal pump is an important mechanical device applied in various industrial fields, and the normal operation of the centrifugal pump is the key to ensuring the normal progress of industrial production. The bearing is one of the core components of the centrifugal pump and is also one of the components with frequent faults in the centrifugal pump. To ensure the safety of the operation of the centrifugal pump, it is extremely important to effectively diagnose the faults of the rolling bearing. With the rapid development of deep learning, intelligent fault diagnosis algorithms using deep learning have been widely used in the field of centrifugal pump bearing fault diagnosis.

[0003] In the prior art, for intelligent fault diagnosis algorithms using deep learning, although good results have been achieved when there is sufficient fault data, in the actual production process of industrial equipment, there is an unbalanced distribution of data used to train the fault diagnosis model, which makes it difficult to build a fault diagnosis model based on deep learning methods. The built fault diagnosis model will have a low recognition accuracy for centrifugal pump bearing faults due to the unbalanced data distribution.

[0004] Therefore, the centrifugal pump fault diagnosis method in the prior art has the problem that the recognition accuracy of centrifugal pump bearing faults is low due to the unbalanced data distribution. Summary of the Invention

[0005] The objective of the embodiments of the present invention is to provide a centrifugal pump fault diagnosis method based on a conditional generative adversarial network, a centrifugal pump fault diagnosis device based on a conditional generative adversarial network, a machine-readable storage medium, and a processor, and this method can improve the recognition accuracy of centrifugal pump bearing faults.

[0006] To achieve the above objective, the first aspect of the present application provides a centrifugal pump fault diagnosis method based on a conditional generative adversarial network, including:

[0007] Obtain the real-time vibration signal of the centrifugal pump bearing;

[0008] Preprocess the real-time vibration signal to obtain the data to be diagnosed;

[0009] Input the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained through the following method:

[0010] Judge the data balance of the sample data to obtain unbalanced data;

[0011] Generate samples for the imbalanced data using the trained conditional generative adversarial network to obtain virtual samples;

[0012] Train a convolutional neural network using the virtual samples and the sample data to obtain a fault diagnosis model.

[0013] In an embodiment of the present application, the sample data includes a fault data set and a healthy data set; the fault diagnosis model is obtained through the following steps:

[0014] Judge the data balance of the fault data set in the sample data to obtain imbalanced data;

[0015] Enhance the imbalanced data using the trained conditional generative adversarial network to obtain a virtual fault data set;

[0016] Construct a balanced data set based on the virtual fault data set, the fault data set, and the healthy data set;

[0017] Train a convolutional neural network according to the balanced data set to obtain a fault diagnosis model.

[0018] In an embodiment of the present application, the step of judging the data balance of the fault data set in the sample data to obtain imbalanced data includes:

[0019] Compare the number of fault data samples in the fault data set with the number of healthy data samples in the healthy data set to obtain a comparison result;

[0020] Judge whether the comparison result meets the preset balance judgment condition;

[0021] If the comparison result meets the preset balance judgment condition, obtain that the data balance judgment result is imbalanced, and use the fault data set as the imbalanced data;

[0022] If the comparison result does not meet the preset balance judgment condition, use a preset data balance judgment model to judge the data balance of the fault data set to obtain imbalanced data.

[0023] In an embodiment of the present application, the step of using a preset data balance judgment model to judge the data balance of the fault data set to obtain imbalanced data includes:

[0024] Calculate the evaluation values of the data of each fault type in the sample data according to the preset data balance judgment model;

[0025] Compare the evaluation values with a preset threshold;

[0026] When the evaluation value is less than the preset threshold, a data balance judgment result of imbalance is obtained, and the corresponding fault type data is used as unbalanced data.

[0027] In the embodiment of the present application, the conditional generative adversarial network is composed of a generator and a discriminator;

[0028] The unbalanced data is enhanced by the trained conditional generative adversarial network to obtain a virtual fault data set, including:

[0029] The conditional generative adversarial network is trained according to the sample data to obtain a trained conditional generative adversarial network;

[0030] The generator in the trained conditional generative adversarial network is used to enhance the unbalanced data to obtain a virtual fault data set.

[0031] In the embodiment of the present application, the training of the conditional generative adversarial network according to the sample data to obtain a trained conditional generative adversarial network includes:

[0032] A1: Randomly generate noise data, and merge the noise data with the data representing category information preset to obtain first input data;

[0033] A2: Take the first input data as the input of the generator, and perform data expansion through a fully connected layer to obtain expanded data;

[0034] A3: Perform multiple upsamplings on the expanded data to obtain virtual data; where each upsampling expands the expanded data to twice its original size;

[0035] A4: Merge the unbalanced data with the data representing category information to obtain second input data, and merge the virtual data with the data representing category information to obtain third input data;

[0036] A5: Input the second input data and the third input data into the discriminator to obtain the probability that the signal is real data;

[0037] A6: Calculate the loss values of the generator and the discriminator respectively according to the objective function, and calculate the gradient backward based on the loss values, and update and adjust the weights of the conditional generative adversarial network according to the gradient;

[0038] A7: Determine whether the probability that the signal is real data and the virtual data meet the preset conditions; if the preset conditions are met, obtain a trained conditional generative adversarial network; if the preset conditions are not met, execute A1 - A6.

[0039] In the embodiments of the present application, the preprocessing of the real-time vibration signal to obtain the data to be diagnosed includes:

[0040] Performing band-pass filtering on the real-time vibration signal to obtain a filtered signal;

[0041] Performing envelope spectrum analysis on the filtered signal to obtain a frequency-domain signal;

[0042] Taking out a plurality of frequency-domain data from the frequency-domain signal to obtain the data to be diagnosed.

[0043] A second aspect of the present application provides a centrifugal pump fault diagnosis device based on a conditional generative adversarial network, including:

[0044] An acquisition module, configured to acquire the real-time vibration signal of the centrifugal pump bearing;

[0045] A preprocessing module, configured to preprocess the real-time vibration signal to obtain the data to be diagnosed;

[0046] A diagnosis module, configured to input the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained by the following method: performing data balance judgment on sample data to obtain unbalanced data; using a trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; using the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model.

[0047] A third aspect of the present application provides a processor configured to execute the above-mentioned centrifugal pump fault diagnosis method based on a conditional generative adversarial network.

[0048] A fourth aspect of the present application provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to cause a machine to execute any one of the above-mentioned centrifugal pump fault diagnosis methods based on a conditional generative adversarial network.

[0049] Through the above technical solution, the real-time vibration signal of the centrifugal pump bearing obtained is preprocessed to obtain the data to be diagnosed; then the data to be diagnosed is input into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained by judging the data balance of sample data to obtain unbalanced data, and using a trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; and then using the virtual samples and the sample data to train a convolutional neural network. By using the conditional generative adversarial network to generate samples for unbalanced data, a large number of virtual samples consistent with the characteristics of actual fault data are obtained; then, the virtual samples and the real sample data are comprehensively used to train the convolutional neural network model, and the obtained fault diagnosis model is applied to the fault diagnosis of the centrifugal pump bearing, so as to achieve accurate identification of the centrifugal pump bearing fault. By judging the data imbalance and then generating samples for the unbalanced data, the enhancement of the unbalanced data is realized, and the problem of low accuracy of centrifugal pump bearing fault identification caused by the unbalanced data distribution in the existing method is effectively solved. In practical applications, by judging the data balance and adopting different process schemes for balanced data and unbalanced data, the speed of model training is greatly improved, and the accuracy of fault diagnosis is also improved.

[0050] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification, and are used together with the following specific implementation manners to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0052] Figure 1 Schematically shows a schematic flowchart of a centrifugal pump fault diagnosis method based on a conditional generative adversarial network according to an embodiment of the present application;

[0053] Figure 2 Schematically shows a flowchart of steps of an intelligent diagnosis method based on a conditional generative adversarial network under data imbalance of a centrifugal pump bearing according to an embodiment of the present application;

[0054] Figure 3 Schematically shows the original signal waveforms of vibration data in four states of the health state, mild outer ring fault, inner ring fault, and severe outer ring fault of a centrifugal pump bearing according to an embodiment of the present application;

[0055] Figure 4 Schematically shows the envelope spectrograms obtained after preprocessing the vibration data in four states of the health state, mild outer ring fault, inner ring fault, and severe outer ring fault of a centrifugal pump bearing according to an embodiment of the present application;

[0056] Figure 5 Schematically shows a virtual sample graph generated by a conditional generative adversarial network based on the imbalance data (outer ring mild fault, inner ring fault) of a centrifugal pump bearing according to an embodiment of the present application;

[0057] Figure 6 Schematically shows a diagram of the classification results of the diagnosis model according to an embodiment of the present application;

[0058] Figure 7 The following schematically shows a structural block diagram of a centrifugal pump fault diagnosis device based on a conditional generative adversarial network according to an embodiment of the present application;

[0059] Figure 8 The internal structure diagram of the device according to the embodiment of the present application is schematically shown.

[0060] Description of Reference Numerals

[0061] 310 - acquisition module; 320 - pre-processing module; 330 - diagnosis module; 40 - device; 401 - processor; 402 - memory; 403 - bus. DETAILED DESCRIPTION

[0062] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0063] Please see Figure 1 and Figure 2 , Figure 1 The following schematically shows a flow chart of a centrifugal pump fault diagnosis method based on a conditional generative adversarial network according to an embodiment of the present application. Figure 2 A flowchart schematically illustrates the steps of an intelligent diagnostic method based on a conditional generative adversarial network (CGN) for centrifugal pump bearing data imbalance, according to an embodiment of the present application. This method collects centrifugal pump bearing vibration data, preprocesses the collected data, and determines the sample balance of each state data. It then uses a CGN to generate samples of the imbalanced data. A convolutional neural network is trained using a combination of virtual and real samples to generate a centrifugal pump bearing fault diagnosis model. Finally, the state samples to be tested are input into the diagnostic model to complete the fault diagnosis. This method enhances the imbalanced data samples, effectively addressing the low accuracy of centrifugal pump bearing fault identification in existing methods with imbalanced data distribution.

[0064] The centrifugal pump fault diagnosis method provided in this embodiment specifically includes the following steps:

[0065] Step 210: Obtain the real-time vibration signal of the centrifugal pump bearing; in this embodiment, the real-time vibration signal of the centrifugal pump bearing can be obtained by acquiring the vibration signal of the centrifugal pump bearing collected by the vibration signal sensor in real time.

[0066] Step 220: Preprocess the real-time vibration signal to obtain the data to be diagnosed; in this embodiment, in order to improve the signal quality, filter out interference components such as noise in the original signal, and extract the periodic features in the signal, the preprocessing includes band-pass filtering the real-time vibration signal and obtaining its envelope spectrum, converting the time-domain signal into a frequency-domain signal, and then taking a certain number of frequency-domain data. Specifically, it includes the following steps:

[0067] First, perform band-pass filtering on the real-time vibration signal to obtain a filtered signal; in this embodiment, the interference signals such as noise in the original vibration signal can be filtered out through band-pass filtering.

[0068] Then, perform envelope spectrum analysis on the filtered signal to obtain a frequency-domain signal; the spectral envelope is a curve formed by connecting the highest amplitude points of different frequencies, which is called the spectral envelope line, in order to obtain the envelope spectrum. The spectrum is a collection of many different frequencies, forming a very wide frequency range, and the amplitudes of different frequencies may be different.

[0069] Finally, take out multiple frequency-domain data from the frequency-domain signal to obtain the data to be diagnosed. For example, the first 1024 data can be extracted as the frequency-domain data. Select multiple frequency-domain data in the envelope spectrum

[0070] By preprocessing the real-time vibration signal, the interference signals such as noise in the vibration signal can be removed, and the periodic features can be extracted through envelope spectrum analysis, so as to improve the accuracy of the feature data for accurate diagnosis in the later stage.

[0071] Step 230: Input the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained through the following method: perform data balance judgment on the sample data to obtain unbalanced data; use the trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; use the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model. By inputting the data to be diagnosed into the preset fault diagnosis model, the fault category of the centrifugal pump bearing can be obtained, and the recognition accuracy of the diagnosis model is as Figure 6 shown.

[0072] It should be noted that the above data balance judgment on the sample data can be a balance judgment on the fault data in the sample data, or a balance judgment on the healthy data in the sample data, or a balance judgment on part of the data in the sample data. This embodiment does not make a limitation.

[0073] In this embodiment, the pre-set fault diagnosis model is a pre-trained fault diagnosis model. The conditional generative adversarial network is used to generate samples for the unbalanced data in the sample data, and a large number of virtual samples consistent with the characteristics of the actual fault data are obtained; then the virtual samples and the real samples are combined to train the convolutional neural network model to obtain the fault diagnosis model, so that the obtained fault diagnosis model eliminates the influence of the unbalanced data.

[0074] Since in the actual production process of industrial equipment, generally the amount of fault data is too small, and the main factor affecting the balance of the sample data is the fault data. Therefore, when making a data balance judgment, generally a balance judgment is made on the fault data. The construction process of the pre-set fault diagnosis model includes the following steps:

[0075] First, obtain sample data, and the sample data includes a fault data set and a healthy data set; in this embodiment, the sample data can be to first use a vibration signal sensor to collect the vibration data of the centrifugal pump bearing, form a bearing sample data set containing a fault data set and a healthy data set, and then preprocess the bearing sample data set to obtain the sample data. The above preprocessing can be to perform band-pass filtering on the bearing sample data set and perform envelope spectrum analysis, convert the bearing time-domain signal into a bearing frequency-domain signal, and take the first 1024 frequency-domain data as a sample.

[0076] For example: a vibration sensor can be installed on the centrifugal pump to collect the vibration data of the centrifugal pump bearing. In this embodiment, the sampling frequency of the vibration sensor can be 25600Hz, and the number of sampling points is 16384, as Figure 3 shown, the collected centrifugal pump bearing data has three fault types: inner ring fault, outer ring mild fault, and outer ring severe fault. The bearings of the three fault types and the fault-free bearings are each collected twice for model training and verification. The above collected data is one-dimensional time-domain data, and the data can be preprocessed in the following manner, and the processed data is as Figure 4 shown:

[0077] 1) Select a one-dimensional vibration signal with a length of 16384 according to the number of sampling points, perform band-pass filtering on the vibration signal, and the frequency range is selected as 6000 - 8000Hz;

[0078] 2) Perform envelope spectrum analysis on the filtered signal, convert the signal into a frequency-domain signal and take its first 1024 points as a data sample.

[0079] Then, perform data balance judgment on the fault data set in the sample data to obtain unbalanced data. In this embodiment, the data sample size and the evaluation value calculated by the data balance evaluation model can be used as the data balance judgment criteria respectively to judge the balance of bearing data. For unbalanced data, perform unbalanced data enhancement, and for balanced data, train the diagnostic model. The specific steps are as follows:

[0080] First step, compare the number of fault data samples in the fault data set with the number of healthy data samples in the healthy data set to obtain a comparison result;

[0081] Second step, judge whether the comparison result meets the preset balance evaluation conditions. In this embodiment, the preset balance evaluation conditions can be set according to actual needs. For example, the preset balance evaluation conditions can be set that the number of fault data samples is less than 5% of the number of healthy data samples.

[0082] Third step, when the comparison result meets the preset balance evaluation conditions, obtain the data balance judgment result as unbalanced, and use the fault data set as unbalanced data;

[0083] Fourth step, when the comparison result does not meet the preset balance evaluation conditions, use the preset data balance evaluation model to perform data balance judgment on the fault data set to obtain unbalanced data.

[0084] Among them, using the preset data balance evaluation model to perform data balance judgment on the fault data set includes the following steps:

[0085] First step, according to the preset data balance evaluation model, calculate the evaluation values of the data of each fault type in the sample data;

[0086] In this embodiment, the data balance evaluation model is a binary classification convolutional neural network. Please refer to Table 1. Table 1 is the network structure table of the data balance evaluation model. Among them, the data balance evaluation model uses fault data as positive examples and healthy data as negative examples, classifies the sample data through the data balance evaluation model, and calculates the evaluation value according to the classification result. Among them, P represents the classification precision rate of fault data, which means the probability that the sample actually is a fault sample among all samples predicted as faults; R represents the classification recall rate of fault data, which means the probability that the sample is predicted as a fault sample among the samples that are actually faults. For balanced samples, both the precision rate and the recall rate should be very high. However, in fact, the above two indicators are contradictory and it is impossible to achieve both high values. The evaluation value F1 value is the harmonic mean of the two, considering both the precision rate and the recall rate at the same time. The evaluation value can be calculated using the following formula:

[0087]

[0088] Among them, P represents the precision rate of fault data classification, which means the probability that the sample actually is a fault sample among all samples predicted as faults; R represents the recall rate of fault data classification, which means the probability that the sample is predicted as a fault sample among the samples that actually are faults.

[0089] Table 1: Network structure table of data balance evaluation model

[0090]

[0091]

[0092] In the second step, compare the evaluation value with a preset threshold; in this embodiment, the preset threshold can be set according to the actual situation. For example, it can be set to 0.6.

[0093] In the third step, when the evaluation value is less than the preset threshold, the data balance judgment result is unbalanced, and the corresponding fault type data is used as unbalanced data.

[0094] For example, taking the number of healthy data samples as the standard, the data samples with the number of fault data samples less than 5% of the number of healthy data samples are determined as unbalanced data; when the number of fault data samples is greater than 5% of the number of healthy data samples, the evaluation value F of the data balance evaluation model is calculated to determine whether it belongs to unbalanced data. That is, taking the fault data as the positive example and the healthy data as the negative example, the evaluation value F1 of each fault type data is calculated using the data balance evaluation model. When the F1 value is greater than 0.6, it indicates data balance, and when it is less than 0.6, it indicates data imbalance.

[0095] By first taking the number of healthy data samples as the standard, comparing the number of fault data samples with the number of healthy data samples to judge the data balance, and then taking the fault data as the positive example and the healthy data as the negative example, calculating the evaluation value of each fault type data using the data balance evaluation model to further judge the data balance, the accuracy of data balance judgment is improved.

[0096] Then, perform data augmentation on the unbalanced data using the trained conditional generative adversarial network to obtain a virtual fault data set; in this embodiment, by using the conditional generative adversarial network to perform data augmentation on the unbalanced data of the centrifugal pump bearing, a large number of virtual fault data samples conforming to the actual fault characteristics can be generated to obtain a virtual fault data set.

[0097] Among them, the conditional generative adversarial network consists of a generator and a discriminator; in this embodiment, the conditional generative adversarial network consists of a generator G and a discriminator D. The role of the generator G is to generate virtual samples as real as possible, and the role of the discriminator D is to correctly distinguish whether the signal is real or virtual as much as possible. Through the confrontation between the generator G and the discriminator D, the conditional generative adversarial network can learn the distribution law of unbalanced data. The generator G in the conditional generative adversarial network structure is an upsampling convolutional neural network, and its structure is shown in Table 2; the discriminator D in the conditional generative adversarial network structure is a binary classification convolutional neural network, and its structure is shown in Table 3.

[0098] Table 2: Structure Table of the Generator

[0099]

[0100] Table 3: Structure Table of the Discriminator

[0101] Layer Name Core Size Number of Cores Core Step Size Padding Normalization Activation Function Conv1 32×1 32 4 0 No LeakyReLU Conv2 5×1 64 4 2 Yes LeakyReLU Conv3 5×1 128 4 2 Yes LeakyReLU Conv4 5×1 256 4 2 Yes LeakyReLU FC / / / / No Sigmoid

[0102] Among them, enhancing the unbalanced data by using the trained conditional generative adversarial network to obtain a virtual fault data set includes the following steps:

[0103] The first step is to train the conditional generative adversarial network according to the sample data to obtain a trained conditional generative adversarial network; in this embodiment, the conditional generative adversarial network can be trained using unbalanced data, and the training process is as follows:

[0104] A1: Randomly generate noise data and merge the noise data with the data representing category information preset to obtain the first input data; in this embodiment, the data representing category information can be preset, for example: for n types of fault data, define them as the first type, the second type... to the nth type, where 1, 2....n are the data representing category information.

[0105] A2: Take the first input data as the input of the generator, and perform data expansion through a fully connected layer to obtain expanded data;

[0106] A3: Perform multiple upsamplings on the expanded data to obtain virtual data; among them, each upsampling expands the expanded data to twice its original size;

[0107] For example: Randomly generate noise data z with a length of 100 that conforms to the standard normal distribution, and merge it with the data c representing category information as the input of the generator G in the conditional generative adversarial network. In the generator G, first expand the data through a fully connected layer; then perform four upsamplings, each time expanding the data to twice its original size, and finally output virtual data G(z|c) with a length of 1024.

[0108] A4: Merge the unbalanced data with the data representing class information to obtain second input data, and merge the virtual data with the data representing class information to obtain third input data;

[0109] A5: Input the second input data and the third input data into the discriminator to obtain the probability that the signal is real data;

[0110] For example: After merging the real data x or the virtual data G(z|c) with the data c representing class information as the input of the discriminator D, after the operation of the discriminator D, finally output the probability that the signal is real data; Here, the real data x refers to the unbalanced data.

[0111] A6: Calculate the loss values of the generator and the discriminator respectively according to the objective function, and calculate the gradients backward based on the loss values, and update and adjust the weights of the conditional generative adversarial network according to the gradients; In this embodiment, calculate the loss values of the generator G and the discriminator D respectively, calculate the gradients backward, and update and adjust the weights of the network.

[0112] Among them, the formula of the objective function V(D,G) is as follows:

[0113]

[0114] Among them, x represents the real sample, z represents the noise data, p data (x) represents the real data distribution, p z (z) represents the noise data distribution, c represents the condition, that is, the sample class information, G(z|c) represents the data generated by the generation model, D(x|c) represents the result judged by the model, E represents the mathematical expectation. For the generator G, the objective function V(D,G) should be minimized, and for the discriminator D, the objective function V(D,G) should be maximized.

[0115] A7: Judge whether the probability that the signal is real data and the virtual data meet the preset conditions; In the case of meeting the preset conditions, obtain the trained conditional generative adversarial network; In the case of not meeting the preset conditions, execute A1-A6. In this embodiment, the preset conditions may be that the virtual samples generated by the generator are consistent with the real samples in characteristics, and the probabilities that the discriminator determines whether the samples are real or virtual are both 50%.

[0116] In the second step, use the generator in the trained conditional generative adversarial network to perform data augmentation on the unbalanced data to obtain a virtual fault data set.

[0117] In this embodiment, steps A1 - A6 are continuously repeated. When the virtual samples generated by the generator are consistent with the real samples in terms of features, and the discriminator determines that the probability of a sample being real or virtual is 50% each, the training of the conditional generative adversarial network is completed. The generated virtual data is as Figure 5 shown.

[0118] Then, based on the virtual fault dataset, fault dataset, and healthy dataset, a balanced dataset is constructed; the conditional generative adversarial network is trained using the imbalanced data. After the training is completed, the obtained bearing data generation model, i.e., the trained conditional generative adversarial network, is used to generate a large number of virtual fault data conforming to the actual fault characteristics, thereby constructing a balanced dataset. By using the conditional generative adversarial network as a sample augmentation method, the classification performance of imbalanced data can be improved.

[0119] Finally, the convolutional neural network is trained based on the balanced dataset to obtain a fault diagnosis model. In this embodiment, please refer to Table 4, which is the structure table of the multi - classification convolutional neural network. The convolutional neural network includes an input layer, convolutional layers, fully - connected layers, and an output layer. The convolutional layers contain different convolutional kernels to implicitly extract sample features; the fully - connected layers synthesize the features, and finally, the fault classification result is output according to the maximum probability.

[0120] Table 4: Structure Table of Multi - classification Convolutional Neural Network

[0121] Layer Name Core Size Number of Cores Core Step Size Padding Normalization Activation Function Conv1 32×1 32 4 0 No ReLU Conv2 5×1 64 4 2 Yes ReLU Conv3 5×1 128 4 2 Yes ReLU Conv4 5×1 256 4 2 Yes ReLU FC / / / / No Softmax

[0122] The above training process can comprehensively use the virtual fault dataset, fault dataset, and healthy dataset as inputs, calculate the data category through the convolutional neural network, calculate the loss between the predicted category of the model and the real category, and update the model parameters through reverse calculation of the gradient. Through continuous training, until the loss between the predicted category and the real category basically remains unchanged, the training is completed, and a fault diagnosis model is obtained.

[0123] In the above implementation, the real-time vibration signals of the centrifugal pump bearings are preprocessed to obtain data to be diagnosed. This data is then input into a pre-set fault diagnosis model to obtain a diagnosis result. The pre-set fault diagnosis model is constructed by determining data balance on sample data to obtain unbalanced data, then generating samples from this unbalanced data using a trained conditional generative adversarial network to obtain virtual samples. A convolutional neural network is then trained using these virtual samples and the sample data. By using the conditional generative adversarial network to generate samples from the unbalanced data, a large number of virtual samples with characteristics consistent with actual fault data are obtained. A convolutional neural network model is then trained by combining the virtual samples with the real sample data. The resulting fault diagnosis model is then applied to centrifugal pump bearing fault diagnosis, achieving accurate identification of centrifugal pump bearing faults. By determining data imbalance and then generating samples from this unbalanced data, the unbalanced data is enhanced, effectively addressing the low accuracy of centrifugal pump bearing fault identification caused by unbalanced data distribution in existing methods. This approach addresses the lack of fault data in the actual production of industrial equipment, ultimately improving fault diagnosis accuracy. In practical applications, by judging the balance of data and adopting different process solutions for balanced data and unbalanced data, the speed of model training is greatly improved, and the accuracy of fault diagnosis is also improved.

[0124] Figure 1 FIG. 1 is a flow chart of a centrifugal pump fault diagnosis method based on a conditional generative adversarial network in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0125] Based on the same inventive concept, this embodiment provides a centrifugal pump fault diagnosis device based on conditional generative adversarial network, please refer to Figure 7 , Figure 7 The schematic diagram shows a structural block diagram of a centrifugal pump fault diagnosis device based on a conditional generative adversarial network according to an embodiment of the present application. The centrifugal pump fault diagnosis device based on a conditional generative adversarial network includes an acquisition module 310, a preprocessing module 320, and a diagnosis module 330, wherein:

[0126] An acquisition module 310 for acquiring real-time vibration signals of a centrifugal pump bearing;

[0127] A preprocessing module 320 for preprocessing the real-time vibration signals to obtain data to be diagnosed;

[0128] A diagnosis module 330 for inputting the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained by the following method: judging data balance for sample data to obtain unbalanced data; using a trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; using the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model.

[0129] The centrifugal pump fault diagnosis device based on a conditional generative adversarial network includes a processor and a memory. The above acquisition module 310, preprocessing module 320, diagnosis module 330, etc. are all stored in the memory as program units, and the processor executes the above program units stored in the memory to implement corresponding functions.

[0130] The processor contains a kernel, and the kernel retrieves the corresponding program units from the memory. One or more kernels can be set, and the fault recognition accuracy of the centrifugal pump bearing can be improved by adjusting the kernel parameters.

[0131] The memory may include non-permanent memory in a computer-readable medium, forms such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0132] An embodiment of the present invention provides a storage medium, on which a program is stored, and when the program is executed by a processor, it implements the centrifugal pump fault diagnosis method based on a conditional generative adversarial network.

[0133] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein when the program runs, it executes the centrifugal pump fault diagnosis method based on a conditional generative adversarial network.

[0134] Please refer to Figure 8 , Figure 8The internal structure diagram of the device according to an embodiment of the present application is schematically shown. An embodiment of the present invention provides a device 40, which includes at least one processor 401, at least one memory 402 connected to the processor 401, and a bus 403; wherein, the processor 401 and the memory 402 communicate with each other through the bus 403; the processor 401 is used to call program instructions in the memory 402 to execute the above-mentioned centrifugal pump fault diagnosis method based on the conditional generative adversarial network. The device 40 herein may be a server, a PC, a PAD, a mobile phone, etc.

[0135] The present application also provides a computer program product, which is suitable for executing a program initialized with the following method steps when executed on a data processing device:

[0136] Obtain the real-time vibration signal of the centrifugal pump bearing;

[0137] Preprocess the real-time vibration signal to obtain the data to be diagnosed;

[0138] Input the data to be diagnosed into a preset fault diagnosis model to obtain a diagnosis result; wherein, the preset fault diagnosis model is obtained through the following steps: judge the data balance of the sample data to obtain imbalance; use the trained conditional generative adversarial network to generate samples for the unbalanced data to obtain virtual samples; use the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model.

[0139] In one embodiment, the construction process of the preset fault diagnosis model includes:

[0140] Obtain sample data, which includes a fault data set and a healthy data set; the fault diagnosis model is obtained through the following steps:

[0141] Use the trained conditional generative adversarial network to perform data augmentation on the unbalanced data to obtain a virtual fault data set;

[0142] Construct a balanced data set according to the virtual fault data set, the fault data set and the healthy data set;

[0143] Train a convolutional neural network according to the balanced data set to obtain a fault diagnosis model.

[0144] In one embodiment, the judgment of the data balance of the fault data set in the sample data to obtain unbalanced data includes:

[0145] Compare the number of fault data samples in the fault data set with the number of healthy data samples in the healthy data set to obtain a comparison result;

[0146] Determine whether the comparison result meets the preset balance evaluation condition;

[0147] When the comparison result meets the preset balance evaluation condition, obtain that the data balance judgment result is unbalanced, and use the fault data set as unbalanced data;

[0148] When the comparison result does not meet the preset balance evaluation condition, use the preset data balance evaluation model to perform data balance judgment on the fault data set, and obtain unbalanced data.

[0149] In one embodiment, the using the preset data balance evaluation model to perform data balance judgment on the fault data set and obtain unbalanced data includes:

[0150] According to the preset data balance evaluation model, calculate the evaluation values of the data of each fault type in the sample data;

[0151] Compare the evaluation value with a preset threshold;

[0152] When the evaluation value is less than the preset threshold, obtain that the data balance judgment result is unbalanced, and use the corresponding fault type data as unbalanced data.

[0153] In one embodiment, the conditional generative adversarial network consists of a generator and a discriminator;

[0154] Using the trained conditional generative adversarial network to perform data augmentation on the unbalanced data to obtain a virtual fault data set includes:

[0155] Train the conditional generative adversarial network according to the sample data to obtain a trained conditional generative adversarial network;

[0156] Use the generator in the trained conditional generative adversarial network to perform data augmentation on the unbalanced data to obtain a virtual fault data set.

[0157] In one embodiment, the training the conditional generative adversarial network according to the sample data to obtain a trained conditional generative adversarial network includes:

[0158] A1: Randomly generate noise data, and merge the noise data with the data representing category information preset to obtain the first input data;

[0159] A2: Use the first input data as the input of the generator, and perform data expansion through a fully connected layer to obtain expanded data;

[0160] A3: Upsample the extended data multiple times to obtain virtual data; wherein, each upsampling expands the extended data to twice its original size.

[0161] A4: Combine the imbalanced data with the data representing class information to obtain second input data, and combine the virtual data with the data representing class information to obtain third input data.

[0162] A5: Input the second input data and the third input data into the discriminator to obtain the probability that the signal is real data.

[0163] A6: Calculate the loss values of the generator and the discriminator respectively according to the objective function, and calculate the gradients backward based on the loss values. Update and adjust the weights of the conditional generative adversarial network according to the gradients.

[0164] A7: Determine whether the probability that the signal is real data and the virtual data meet the preset conditions; if the preset conditions are met, obtain the trained conditional generative adversarial network; if the preset conditions are not met, execute A1 - A6.

[0165] In one embodiment, the preprocessing of the real - time vibration signal to obtain the data to be diagnosed includes:

[0166] Perform band - pass filtering on the real - time vibration signal to obtain a filtered signal.

[0167] Perform envelope spectrum analysis on the filtered signal to obtain a frequency - domain signal.

[0168] Extract multiple frequency - domain data from the frequency - domain signal to obtain the data to be diagnosed.

[0169] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer - usable storage media (including but not limited to disk memories, CD - ROMs, optical memories, etc.) containing computer - usable program codes.

[0170] This application is described by reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one or more flows and / or blocks. Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0171] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more flows and / or blocks. Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0172] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows and / or blocks. Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one or more blocks.

[0173] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0174] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0175] A computer-readable medium includes both permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0176] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0177] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A centrifugal pump fault diagnosis method based on conditional generative adversarial network, characterized in that: include: Obtain real-time vibration signals of centrifugal pump bearings; Preprocessing the real-time vibration signal to obtain data to be diagnosed; The data to be diagnosed is input into a preset fault diagnosis model to obtain a diagnosis result; wherein the fault diagnosis model is obtained by: Perform data balance judgment on the sample data to obtain unbalanced data; Generate samples from the imbalanced data using a trained conditional generative adversarial network to obtain virtual samples; Using the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model; The sample data includes a fault data set and a healthy data set; the fault diagnosis model is obtained by: Performing data balance judgment on the fault data set in the sample data to obtain unbalanced data; The unbalanced data is enhanced using a trained conditional generative adversarial network to obtain a virtual fault data set; Constructing a balanced data set based on the virtual fault data set, the fault data set and the healthy data set; Training a convolutional neural network based on the balanced data set to obtain a fault diagnosis model; The conditional generative adversarial network is composed of a generator and a discriminator; The unbalanced data is enhanced using a trained conditional generative adversarial network to obtain a virtual fault dataset, including: Training a conditional generative adversarial network based on the sample data to obtain a trained conditional generative adversarial network; Using the generator in the trained conditional generative adversarial network to perform data augmentation on the imbalanced data to obtain a virtual fault data set; The step of training the conditional generative adversarial network according to the sample data to obtain a trained conditional generative adversarial network includes: A1: randomly generating noise data, and combining the noise data with preset data representing category information to obtain first input data; A2: The first input data is used as the input of the generator, and data expansion is performed through a fully connected layer to obtain expanded data; A3: performing multiple upsampling on the extended data to obtain virtual data; wherein each upsampling expands the extended data to twice the original value; A4: Merge the unbalanced data with the data representing the category information to obtain second input data, and merge the dummy data with the data representing the category information to obtain third input data; A5: Input the second input data and the third input data into the discriminator to obtain the probability that the signal is true data; A6: Calculate the loss values of the generator and the discriminator respectively according to the objective function, and reversely calculate the gradients based on the loss values, and update and adjust the weights of the conditional generative adversarial network according to the gradients; A7: Determine the probability that the signal is real data and whether the virtual data meets preset conditions; if the preset conditions are met, obtain a trained conditional generative adversarial network; if the preset conditions are not met, execute A1-A6.

2. The centrifugal pump fault diagnosis method based on conditional generative adversarial network according to claim 1 is characterized in that: The performing data balance judgment on the fault data set in the sample data to obtain unbalanced data includes: Comparing the number of fault data samples in the fault data set with the number of healthy data samples in the healthy data set to obtain a comparison result; Determining whether the comparison result satisfies a preset balance evaluation condition; When the comparison result satisfies a preset balance judgment condition, a data balance judgment result is obtained as unbalanced, and the fault data set is used as unbalanced data; In the case that the comparison result does not satisfy the preset balance judgment condition, a preset data balance judgment model is used to perform data balance judgment on the fault data set to obtain unbalanced data.

3. The centrifugal pump fault diagnosis method based on conditional generative adversarial network according to claim 2 is characterized in that: The method of using a preset data balance judgment model to perform data balance judgment on the fault data set to obtain unbalanced data includes: Calculate the evaluation value of each fault type data in the sample data according to the preset data balance evaluation model; Comparing the evaluation value with a preset threshold; In the case that the evaluation value is less than the preset threshold value, the data balance judgment result is obtained as unbalanced, and the corresponding fault type data is used as unbalanced data.

4. The centrifugal pump fault diagnosis method based on conditional generative adversarial network according to claim 1 is characterized in that: The preprocessing of the real-time vibration signal to obtain data to be diagnosed includes: performing bandpass filtering on the real-time vibration signal to obtain a filtered signal; Performing envelope spectrum analysis on the filtered signal to obtain a frequency domain signal; A plurality of frequency domain data are extracted from the frequency domain signal to obtain data to be diagnosed.

5. A centrifugal pump fault diagnosis device based on conditional generative adversarial network, characterized in that: include: An acquisition module, used for acquiring real-time vibration signals of centrifugal pump bearings; A preprocessing module, used for preprocessing the real-time vibration signal to obtain data to be diagnosed; A diagnostic module is used to input the data to be diagnosed into a preset fault diagnosis model to obtain a diagnostic result; wherein, the fault diagnosis model is obtained by: performing a data balance judgment on the sample data to obtain unbalanced data; using a trained conditional generative adversarial network to generate samples of the unbalanced data to obtain virtual samples; using the virtual samples and the sample data to train a convolutional neural network to obtain a fault diagnosis model; wherein, the sample data includes a fault data set and a healthy data set; the fault diagnosis model is obtained by: performing a data balance judgment on the fault data set in the sample data to obtain unbalanced data; sampling the unbalanced data The method comprises the following steps: using a trained conditional generative adversarial network to perform data enhancement to obtain a virtual fault data set; constructing a balanced data set based on the virtual fault data set, the fault data set and the healthy data set; training a convolutional neural network based on the balanced data set to obtain a fault diagnosis model; wherein the conditional generative adversarial network is composed of a generator and a discriminator; performing data enhancement on the unbalanced data using the trained conditional generative adversarial network to obtain a virtual fault data set, including: training the conditional generative adversarial network based on the sample data to obtain a trained conditional generative adversarial network; and performing data enhancement on the unbalanced data using the generator in the trained conditional generative adversarial network. Enhancement is performed to obtain a virtual fault data set; wherein, the conditional generative adversarial network is trained according to the sample data to obtain a trained conditional generative adversarial network, including: A1: randomly generating noise data, and merging the noise data with preset data representing category information to obtain first input data; A2: using the first input data as the input of the generator, performing data expansion through a fully connected layer to obtain expanded data; A3: performing multiple upsampling on the expanded data to obtain virtual data; wherein each upsampling expands the expanded data to twice the original value; A4: merging the unbalanced data with the data representing category information to obtain a second input data. A5: input the second input data and the third input data into the discriminator to obtain the probability that the signal is real data; A6: calculate the loss values of the generator and the discriminator respectively according to the objective function, and reversely calculate the gradient based on the loss value, and adjust the weight of the conditional generative adversarial network according to the gradient update; A7: judge whether the probability that the signal is real data and the virtual data meet the preset conditions; if the preset conditions are met, obtain the trained conditional generative adversarial network; if the preset conditions are not met, execute A1-A6.

6. A processor, characterized in that: The method is configured to execute the centrifugal pump fault diagnosis method based on conditional generative adversarial network according to any one of claims 1 to 4.

7. A machine-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the centrifugal pump fault diagnosis method based on conditional generative adversarial network according to any one of claims 1 to 4.

Images