Pump valve fault diagnosis method and device

Abstract

The present application proposes a pump valve fault diagnosis method and device, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum, so as to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a pump valve fault diagnosis method and device. Background Art

[0002] Underground coal mining is a harsh environment with complex and ever-changing operating conditions. Reciprocating piston pumps operating continuously for long periods of time are prone to frequent failures. Suction and discharge valve failure is a typical example of this type of failure, which not only impacts coal mining efficiency but can also cause severe economic losses and safety hazards. Therefore, condition monitoring and fault diagnosis of pumps and valves are crucial for the safe and stable operation of equipment.

[0003] In the related art, sensors are installed on the valves of mining reciprocating plunger pumps to collect equipment status information, and then equipment abnormality warning and fault diagnosis are realized through data analysis to prevent the occurrence of serious faults and reduce economic losses. Based on a large amount of different state data of pumps and valves with similar data volume, a pump and valve fault diagnosis model is established in combination with data-driven technologies such as signal processing, machine learning or deep learning to realize the diagnosis of pump and valve faults. However, the state monitoring data of the suction and discharge valves of the reciprocating plunger pumps in the related art obeys a long-tail distribution. The sample size under normal conditions is very large, while the sample size of faults is very small, resulting in a data category imbalance problem that must be paid attention to in engineering. The related pump and valve fault diagnosis method is proposed under the assumption of data balance, which results in the above-mentioned pump and valve fault diagnosis method not having generalization ability. Summary of the Invention

[0004] The present application provides a pump valve fault diagnosis method and device.

[0005] The first aspect of the present application provides a pump valve fault diagnosis method, which includes: obtaining a full-cycle signal sample of the pump valve and a pump valve state label corresponding to the full-cycle signal sample, and performing signal transformation on the full-cycle signal sample to obtain a signal spectrum corresponding to the full-cycle signal sample; inputting the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum; training the conditional variational autoencoder based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution to determine the model parameters of the conditional variational autoencoder; constructing the decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and using the decoder model as the pump valve fault diagnosis model; merging the real-time signal spectrum corresponding to the real-time full-cycle signal sample of the pump valve and the feature distribution and inputting them into the fault diagnosis model to obtain a real-time state prediction label of the pump valve; and determining whether the pump valve is faulty based on the real-time state prediction label of the pump valve.

[0006] In one embodiment of the present application, the obtaining of the whole-cycle signal sample of the pump valve and the pump-valve status label corresponding to the whole-cycle signal sample, and performing fast Fourier transform on the whole-cycle signal sample to obtain the signal spectrum corresponding to the whole-cycle signal sample, includes: obtaining the vibration signal corresponding to the pump valve, and the key phase signal of the crankshaft corresponding to the pump valve; windowing the vibration signal and the key phase signal to obtain the whole-cycle signal sample of the pump valve; state-labeling the pump-valve status corresponding to the whole-cycle signal sample to obtain the pump-valve status label of the pump valve; performing fast Fourier transform on the whole-cycle signal sample to obtain the initial signal spectrum corresponding to the whole-cycle signal sample; and normalizing the initial signal spectrum to obtain a normalized signal spectrum.

[0007] In one embodiment of the present application, the conditional variational autoencoder includes an encoder, a latent feature space representation layer and a decoder, and the signal spectrum is input into the conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum, including: inputting the signal spectrum into the encoder to obtain a feature vector corresponding to the signal spectrum; inputting the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector; and inputting the feature distribution and the signal spectrum into the decoder to obtain a predicted state label corresponding to the signal spectrum.

[0008] In one embodiment of the present application, inputting the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector includes: inputting the feature vector into the latent feature space representation layer to obtain the mean and variance of the feature vector; and randomly sampling the mean and variance to obtain the feature distribution corresponding to the feature vector.

[0009] In one embodiment of the present application, the conditional variational autoencoder is trained based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution to determine the model parameters of the conditional variational autoencoder, including: using the pump valve state label as the input of the conditional variational autoencoder and the predicted state label as the output of the conditional variational autoencoder to generate an initial model of the conditional variational autoencoder; based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, training the initial model until the initial model meets the preset termination criteria to obtain the target model of the conditional variational autoencoder; and using the target model parameters in the target model as the model parameters of the conditional variational autoencoder.

[0010] The present application proposes a pump valve fault diagnosis method, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum, so as to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

[0011] The third aspect of the present application provides a pump valve fault diagnosis device, which includes: an acquisition module for acquiring a full-cycle signal sample of the pump valve and a pump valve state label corresponding to the full-cycle signal sample, and performing signal transformation on the full-cycle signal sample to obtain a signal spectrum corresponding to the full-cycle signal sample; a first generation module for inputting the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum; a first determination module for training the conditional variational autoencoder based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and a preset data distribution, and determining the model parameters of the conditional variational autoencoder; a construction module for constructing the decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and using the decoder model as the fault diagnosis model of the pump valve; a second generation module for merging the real-time signal spectrum corresponding to the real-time full-cycle signal sample of the pump valve and the feature distribution and inputting them into the fault diagnosis model to obtain a real-time state prediction label of the pump valve; and a second determination module for determining whether the pump valve is faulty based on the real-time state prediction label of the pump valve.

[0012] In one embodiment of the present application, the acquisition module is specifically used to: acquire the vibration signal corresponding to the pump valve, and the key phase signal of the crankshaft corresponding to the pump valve; perform windowing processing on the vibration signal and the key phase signal to obtain a full-cycle signal sample of the pump valve; perform state labeling on the pump valve state corresponding to the full-cycle signal sample to obtain a pump valve state label of the pump valve; perform fast Fourier transform on the full-cycle signal sample to obtain an initial signal spectrum corresponding to the full-cycle signal sample; and perform normalization processing on the initial signal spectrum to obtain a normalized signal spectrum.

[0013] In one embodiment of the present application, the conditional variational autoencoder includes an encoder, a latent feature space representation layer and a decoder, and the first generation module includes: an input unit for inputting the signal spectrum into the encoder to obtain a feature vector corresponding to the signal spectrum; a sampling unit for inputting the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector; and a generation unit for inputting the feature distribution and the signal spectrum into the decoder to obtain a predicted state label corresponding to the signal spectrum.

[0014] In one embodiment of the present application, the sampling unit is specifically used to: input the feature vector into the latent feature space representation layer to obtain the mean and variance of the feature vector; and perform random distribution sampling on the mean and variance to obtain the feature distribution corresponding to the feature vector.

[0015] In one embodiment of the present application, the first determination module is specifically used to: generate an initial model of the conditional variational autoencoder based on the pump valve state label as the input of the conditional variational autoencoder and the predicted state label as the output of the conditional variational autoencoder; train the initial model based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution until the initial model meets the preset termination criterion to obtain the target model of the conditional variational autoencoder; and use the target model parameters in the target model as the model parameters of the conditional variational autoencoder.

[0016] The present application proposes a pump valve fault diagnosis device, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

[0017] The fifth aspect of the present application provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the pump valve fault diagnosis method in the embodiment of the present application.

[0018] The sixth aspect of the present application provides a computer program product, which implements the pump valve fault diagnosis method in the embodiment of the present application when the instruction processor in the computer program product is executed.

[0019] Other effects of the above optional manner will be described below in conjunction with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a pump valve fault diagnosis method provided by an embodiment of the present application;

[0021] Figure 2 1 is a flow chart of another pump valve fault diagnosis method provided in an embodiment of the present application;

[0022] Figure 3 1 is a flow chart of another pump valve fault diagnosis method provided in an embodiment of the present application;

[0023] Figure 4 This is a training diagram of a conditional variational autoencoder provided in an embodiment of the present application;

[0024] Figure 5 This is a confusion logic matrix diagram of a pump valve status diagnosis result provided by an embodiment of the present application;

[0025] Figure 6 It is a structural schematic diagram of a pump valve fault diagnosis device provided in an embodiment of the present application;

[0026] Figure 7 It is a structural schematic diagram of another pump valve fault diagnosis device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0028] The pump valve fault diagnosis method, device and electronic device according to the embodiments of the present application are described below with reference to the accompanying drawings.

[0029] Figure 1The figure is a flow chart of a pump and valve fault diagnosis method provided in an embodiment of the present application. It should be noted that the method provided in this embodiment is performed by a pump and valve fault diagnosis device, which can be implemented in software and / or hardware. The pump and valve fault diagnosis device in this embodiment can be configured in an electronic device, which can include a server. This embodiment does not specifically limit the electronic device.

[0030] Figure 1 It is a flow chart of a pump valve fault diagnosis method provided in an embodiment of the present application.

[0031] like Figure 1 As shown, the pump valve fault diagnosis method may include:

[0032] Step 101 : obtaining a full-cycle signal sample of a pump valve and a pump valve state label corresponding to the full-cycle signal sample, and performing signal transformation on the full-cycle signal sample to obtain a signal spectrum corresponding to the full-cycle signal sample.

[0033] In some embodiments, the pump valve may be a discharge valve and a suction valve of a reciprocating plunger pump in an underground coal mining process, but is not limited thereto and this embodiment does not make any specific limitation thereto.

[0034] In some embodiments, the pump valve status label can be divided into multiple types, including a pump valve fault label, a pump valve abnormality label, and a pump valve normal label, but is not limited thereto.

[0035] Specifically, the discharge valve fault label, the suction valve fault label, or both the discharge and suction valves are faulty labels, and the discharge valve normal label, the suction valve normal label, or both the discharge and suction valves are normal labels, which are not specifically limited in this embodiment.

[0036] It can be understood that the full-cycle signal samples of the pump valve can be controlled to be compared with the standard full-cycle signal samples of the normal operation of the pump valve to realize abnormal diagnosis of the full-cycle signal samples, so as to determine the pump valve status label corresponding to the full-cycle signal sample and realize accurate labeling of the pump valve status label corresponding to the full-cycle signal sample.

[0037] Among them, the signal spectrum corresponding to the full-cycle signal samples of the pump valve has less abnormal signal spectrum data, and the signal spectrum when the pump valve is operating normally has more.

[0038] In some embodiments, the signal transformation performed on the integer-period signal samples may be a Fourier transform. Specifically, the integer-period signal samples may be subjected to a fast Fourier transform to obtain a signal spectrum corresponding to the integer-period signal samples, but is not limited thereto.

[0039] Step 102: Input the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum.

[0040] In some embodiments, the conditional variational autoencoder may include two parts, an encoder and a decoder, and the encoder and the decoder may be modeled using a multi-layer perceptron, wherein the encoder and the decoder are connected through a latent feature space representation layer of the data. Thus, based on the multi-layer perceptron and the conditional variational autoencoder, accurate detection of the unbalanced signal spectrum is achieved.

[0041] Step 103 : Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the conditional variational autoencoder is trained to determine the model parameters of the conditional variational autoencoder.

[0042] In some embodiments, a conditional variational autoencoder is trained based on the difference between the pump valve state label and the predicted state label. One implementation method for determining the model parameters of the conditional variational autoencoder can be to use the pump valve state label as the input of the conditional variational autoencoder and the predicted state label as the output of the conditional variational autoencoder to generate an initial model of the conditional variational autoencoder. Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the initial model is trained until the initial model meets the preset termination criteria to obtain the target model of the conditional variational autoencoder. The target model parameters in the target model are used as the model parameters of the conditional variational autoencoder, thereby obtaining the optimal model parameters of the conditional variational autoencoder and improving the accuracy of the fault diagnosis model.

[0043] Specifically, after obtaining the initial model of the conditional variational autoencoder, the initial model can be optimized and trained by setting the learning rate and selecting an optimization trainer so that the initial model meets the preset termination criteria to obtain the target model of the conditional variational autoencoder, and the target model parameters in the target model are used as the model parameters of the conditional variational autoencoder.

[0044] The set learning rate may be set to 1e-4, or may be set by relevant technical personnel according to actual business scenarios, and this embodiment does not specifically limit this.

[0045] The optimization trainer may include but is not limited to Adam.

[0046] The preset termination criterion may be that the initial model training is terminated when the loss value stops decreasing, but is not limited thereto.

[0047] It can be understood that the loss function corresponding to the loss value of the initial model training may include but is not limited to using the focal loss function to measure the difference between the pump valve state label and the predicted state label, and using the Kullback-Leible divergence to measure the difference between the potential feature space distribution and the standard normal distribution. This embodiment does not make specific limitations on this.

[0048] Specifically, taking the focal loss function to measure the difference between the pump valve state label and the predicted state label, and the Kullback-Leible divergence to measure the difference between the potential feature space distribution and the standard normal distribution as an example, the definition formula is as follows.

[0049]

[0050] Among them, the balance parameter α is used to deal with the imbalance problem between the number of normal and fault samples, P t Represents the label prediction probability of the sample, and the focus parameter γ generally takes 2 values, μ and σ 2 represents the mean and variance, d represents the number of sampled variables in the latent space, and its value is 512.

[0051] In other embodiments, the optimization goal may be to minimize the difference between the pump valve state label and the predicted state label, thereby determining the accurate model parameters of the conditional variational autoencoder.

[0052] Step 104: construct a decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve.

[0053] In some embodiments, the decoder in the trained conditional variational autoencoder can be directly used as a fault diagnosis model for the pump valve. It is also possible to construct a decoder model based on the model parameters of the decoder in the model parameters of the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve.

[0054] Step 105 : The real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump and valve are combined and input into the fault diagnosis model to obtain a real-time state prediction label of the pump and valve.

[0055] In some embodiments, the influence of the state labels of different real-time full-cycle signal samples on the fault diagnosis model is balanced by feature distribution, so as to accurately diagnose the real-time signal spectrum corresponding to the real-time full-cycle signal sample, so as to obtain a more accurate real-time state prediction label of the pump and valve, and realize accurate diagnosis of the pump and valve state.

[0056] Step 106 : Determine whether the pump or valve is faulty based on the real-time status prediction tag of the pump or valve.

[0057] In some embodiments, when the real-time state prediction label of the pump valve is a normal pump valve label, it indicates that the pump valve is operating in a normal state. When the real-time state prediction label of the pump valve is an abnormal pump valve label, it indicates that the pump valve is in an abnormal state. The pump valve can be manually regulated based on the abnormal pump valve label to eliminate faults and ensure the safe and stable operation of the pump valve.

[0058] The present application proposes a pump valve fault diagnosis method, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum, so as to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

[0059] Figure 2 It is a flow chart of another pump valve fault diagnosis method provided in an embodiment of the present application.

[0060] Step 201: Acquire a vibration signal corresponding to a pump valve and a key phase signal of a crankshaft corresponding to the pump valve.

[0061] In some embodiments, the vibration signal corresponding to the pump valve may be measured by a vibration sensor installed on the pump valve, but is not limited thereto.

[0062] Specifically, taking the plunger pump discharge valve as an example, the state of the pump valve can be sensed by a vibration sensor installed on the sealing plate of the plunger pump discharge chamber, and the state information of the pump valve can be obtained, and the vibration signal corresponding to the pump valve can be obtained from the state information.

[0063] In some embodiments, an implementation method for obtaining the key phase signal of the pump valve corresponding to the crankshaft may be to obtain the key phase signal of the crankshaft through a signal sensing head installed on the crankshaft and a speed sensor on the crankcase end cover, but is not limited thereto.

[0064] Step 202: Perform windowing processing on the vibration signal and the key phase signal to obtain full-cycle signal samples of the pump valve.

[0065] In some embodiments, in order to reduce the impact of noise on the full-cycle signal samples of the pump valve, the vibration signal and the key phase signal can be windowed by a window function to obtain the full-cycle signal samples of the pump valve, thereby avoiding noise interference while ensuring the continuity of the full-cycle signal samples of the pump valve.

[0066] Step 203 : labeling the pump valve status corresponding to the full cycle signal sample to obtain the pump valve status label of the pump valve.

[0067] In some embodiments, the pump valve states corresponding to the full-cycle signal samples can be manually labeled to obtain the pump valve state label of the pump valve. The pump valve states corresponding to the historical full-cycle signal samples can also be model trained to obtain a pump valve state labeling model. The pump valve states corresponding to the full-cycle signal samples can be labeled using the pump valve state labeling model to obtain the pump valve state label of the pump valve. This embodiment does not make any specific restrictions on this.

[0068] Step 204 : Perform fast Fourier transform on the integral period signal samples to obtain an initial signal spectrum corresponding to the integral period signal samples.

[0069] In some embodiments, in order to improve the processing efficiency of the integer-cycle signal samples, the integer-cycle signal samples can be subjected to a fast Fourier transform to obtain the initial signal spectrum corresponding to the integer-cycle signal samples, that is, to convert the time signal into a frequency domain signal that is easy to analyze, thereby improving the processing efficiency of the integer-cycle signal samples.

[0070] Step 205 : normalize the initial signal spectrum to obtain a normalized signal spectrum.

[0071] In some embodiments, the initial signal spectrum may be normalized by a linear function method, a logarithmic function method, or an inverse cotangent function method to obtain a normalized signal spectrum, but the present invention is not limited thereto and this embodiment does not make any specific limitation thereto.

[0072] Step 206: Input the signal spectrum into the conditional variational autoencoder to obtain the feature distribution and predicted state label corresponding to the signal spectrum.

[0073] Step 207 : Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the conditional variational autoencoder is trained to determine the model parameters of the conditional variational autoencoder.

[0074] Step 208: construct a decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve.

[0075] Step 209 : The real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump and valve are combined and input into the fault diagnosis model to obtain a real-time state prediction label of the pump and valve.

[0076] Step 210: Determine whether the pump or valve is faulty based on the real-time status prediction tag of the pump or valve.

[0077] It should be noted that, for the specific implementation of steps 206 to 210, reference may be made to the relevant description in the above embodiment.

[0078] The present application proposes a pump valve fault diagnosis method, which obtains a vibration signal corresponding to the pump valve and a key phase signal of a crankshaft corresponding to the pump valve, performs windowing processing on the vibration signal and the key phase signal to obtain a full-cycle signal sample of the pump valve, performs state labeling on the pump valve state corresponding to the full-cycle signal sample to obtain a pump valve state label of the pump valve, performs fast Fourier transform on the full-cycle signal sample to obtain an initial signal spectrum corresponding to the full-cycle signal sample, normalizes the initial signal spectrum to obtain a normalized signal spectrum, inputs the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum, and according to the difference between the pump valve state label and the predicted state label and the feature distribution, The difference between the characteristic distribution and the preset data distribution is determined, and the conditional variational autoencoder is trained to determine the model parameters of the conditional variational autoencoder. Based on the model parameters of the decoder in the conditional variational autoencoder, a decoder model is constructed, and used as a fault diagnosis model for the pump valve. The real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve, so as to determine whether the pump valve is faulty. Therefore, in the fault diagnosis process of the pump valve, based on the signal spectrum corresponding to the accurate pump valve full-cycle signal samples and the pump valve state label, the fault diagnosis model constructed by the characteristic distribution of the signal spectrum is accurately obtained, thereby realizing accurate diagnosis of the pump valve state.

[0079] Figure 3 This is a flow chart of another pump valve fault diagnosis method provided in an embodiment of the present application, wherein the conditional variational autoencoder includes an encoder, a latent feature space representation layer and a decoder.

[0080] Step 301 : Acquire the full-cycle signal samples of the pump valve and the pump valve state labels corresponding to the full-cycle signal samples, and perform signal transformation on the full-cycle signal samples to obtain the signal spectrum corresponding to the full-cycle signal samples.

[0081] It should be noted that, for the specific implementation of step 301, reference may be made to the relevant description in the above embodiment.

[0082] Step 302: Input the signal spectrum into an encoder to obtain a feature vector corresponding to the signal spectrum.

[0083] In some embodiments, the encoder can be constructed by a one-dimensional convolutional neural network or a multi-layer perceptron, but is not limited thereto and this embodiment does not make any specific limitations on this.

[0084] In other embodiments, multiple hidden layers may be provided in the encoder to enhance the data feature extraction capability of the encoder, thereby improving the accuracy of the feature vector.

[0085] Specifically, the signal spectrum can be input into the encoder, and the pump valve state label corresponding to the full cycle signal sample can also be input into the encoder to perform feature extraction on the signal spectrum to obtain a feature vector corresponding to the signal spectrum.

[0086] Specifically, if the dimension of the signal spectrum is 1024 and there are 4 categories of pump valve status labels, the input dimension of the encoder can be set to 1028, and the number of neurons in the network hidden layer can be set to 2048 and 1024 to achieve feature extraction of the signal spectrum.

[0087] Step 303: Input the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector.

[0088] In some embodiments, an implementation method of inputting a feature vector into a latent feature space representation layer to sample a feature distribution corresponding to the feature vector may be to input the feature vector into the latent feature space representation layer to obtain the mean and variance of the feature vector, and randomly sample the mean and variance to obtain the feature distribution corresponding to the feature vector.

[0089] Among them, for the mean and variance of the potential feature space representation layer, 512 neurons can be set for sampling.

[0090] Step 304: Input the feature distribution and the signal spectrum into a decoder to obtain a predicted state label corresponding to the signal spectrum.

[0091] In some embodiments, the decoder may be constructed from multiple hidden layers, but is not limited thereto.

[0092] Specifically, the ability to restore data from the signal spectrum can be enhanced through multiple hidden layers, and the pump and valve state labels corresponding to the full-cycle signal samples can be balanced in combination with the feature distribution to obtain the predicted state labels corresponding to the signal spectrum.

[0093] For example, when the mean and variance of the latent feature space representation layer are both set to contain 512 neurons, the input dimension of the decoder can be set to 1536 to output the predicted state label corresponding to the signal spectrum.

[0094] Step 305 : Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the conditional variational autoencoder is trained to determine the model parameters of the conditional variational autoencoder.

[0095] Step 306: construct a decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve.

[0096] Step 307 : The real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump and valve are combined and input into the fault diagnosis model to obtain a real-time state prediction label of the pump and valve.

[0097] Step 308: Determine whether the pump or valve is faulty based on the real-time status prediction tag of the pump or valve.

[0098] The present application proposes a pump valve fault diagnosis method, which obtains full-cycle signal samples of the pump valve and their corresponding pump valve state labels, and obtains a signal spectrum corresponding to the full-cycle signal samples. The signal spectrum is input into an encoder to obtain a feature vector corresponding to the signal spectrum. The feature vector is input into a latent feature space representation layer to sample a feature distribution corresponding to the feature vector. The feature distribution and the signal spectrum are input into a decoder to obtain a predicted state label corresponding to the signal spectrum. Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and a preset data distribution, a conditional variational autoencoder is trained to determine the model parameters of the conditional variational autoencoder. Based on the model parameters of the decoder in the conditional variational autoencoder, a decoder model is constructed and used as a pump valve fault diagnosis model. The real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal samples of the pump valve are combined and input into the fault diagnosis model to obtain a real-time state prediction label of the pump valve, thereby determining whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum, achieving accurate diagnosis of the pump valve state, and improving the universality and accuracy of the fault diagnosis model.

[0099] In summary, this application also proposes a framework diagram of a conditional variational autoencoder, such as Figure 4As shown, the conditional variational autoencoder consists of an encoder, a latent feature space representation layer and a decoder. Specifically, the signal spectrum corresponding to the full-cycle signal sample of the pump valve and the pump valve state label corresponding to the full-cycle signal sample can be input into the encoder for encoding to obtain the feature vector corresponding to the signal spectrum, and then the feature vector is input into the latent feature space representation layer to obtain the mean and variance corresponding to the feature vector, and then the mean and variance are resampled according to the preset data distribution to obtain the sampling variable z of the feature distribution resampled corresponding to the feature vector, so that the sampling variable and the signal spectrum are merged and input into the decoder to obtain the predicted state label corresponding to the signal spectrum. Thus, based on the difference between the predicted state label and the pump valve state label, the conditional variational autoencoder is trained to improve the accuracy of the conditional variational autoencoder.

[0100] In addition, it can be understood that, taking the fault diagnosis of the discharge valve and suction valve of the reciprocating plunger pump in the process of underground coal mining as an example, Figure 5 As shown, Figure 5 It is a confusion matrix diagram of a pump valve status diagnosis result provided by an embodiment of the present application. Specifically, the real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump valve are merged and input into the fault diagnosis model to obtain the real-time status prediction label of the pump valve. The pump valve status label is used as the vertical coordinate and the pump valve real-time status prediction label is used as the horizontal coordinate. The four label types of the pump valve status label and the pump valve real-time status prediction label, namely normal, discharge valve fault, suction valve fault and both suction and discharge valve fault, are used for matrix division to obtain a confusion matrix diagram of the pump valve status diagnosis result.

[0101] Figure 6 It is a structural schematic diagram of a pump valve fault diagnosis device provided in an embodiment of the present application.

[0102] like Figure 6 As shown, the pump valve fault diagnosis device 600 includes:

[0103] The acquisition module 601 is used to acquire the full-cycle signal samples of the pump valve and the pump valve state labels corresponding to the full-cycle signal samples, and perform signal transformation on the full-cycle signal samples to obtain the signal spectrum corresponding to the full-cycle signal samples.

[0104] The first generation module 602 is used to input the signal spectrum into the conditional variational autoencoder to obtain the feature distribution and predicted state label corresponding to the signal spectrum.

[0105] The first determination module 603 is used to train the conditional variational autoencoder according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and determine the model parameters of the conditional variational autoencoder.

[0106] The construction module 604 is used to construct a decoder model based on the model parameters of the decoder in the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve.

[0107] The second generating module 605 is used to combine the real-time signal spectrum and characteristic distribution corresponding to the real-time full-cycle signal samples of the pump valve and input them into the fault diagnosis model to obtain the real-time state prediction label of the pump valve.

[0108] The second determination module 606 is configured to determine whether the pump valve is faulty based on the real-time status prediction tag of the pump valve.

[0109] The present application proposes a pump valve fault diagnosis device, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

[0110] In one embodiment of the present application, Figure 7 As shown, the acquisition module 601 is specifically used to:

[0111] Obtain the vibration signal corresponding to the pump valve, as well as the key phase signal of the crankshaft corresponding to the pump valve.

[0112] The vibration signal and key phase signal are windowed to obtain the full cycle signal samples of the pump valve.

[0113] The pump valve states corresponding to the full cycle signal samples are labeled to obtain the pump valve state labels of the pump valves.

[0114] Perform fast Fourier transform on the integer period signal samples to obtain the initial signal spectrum corresponding to the integer period signal samples.

[0115] The initial signal spectrum is normalized to obtain a normalized signal spectrum.

[0116] In one embodiment of the present application, Figure 7 As shown, the conditional variational autoencoder includes an encoder, a potential feature space representation layer and a decoder, and the first generation module 602 includes:

[0117] The input unit 6021 is used to input the signal spectrum into the encoder to obtain a feature vector corresponding to the signal spectrum.

[0118] The sampling unit 6022 is used to input the feature vector into the potential feature space representation layer to sample the feature distribution corresponding to the feature vector.

[0119] 6023 is used to input the feature distribution and the signal spectrum into the decoder to obtain the predicted state label corresponding to the signal spectrum.

[0120] In one embodiment of the present application, Figure 7 As shown, the sampling unit 6022 is specifically used to:

[0121] The feature vector is input into the latent feature space representation layer to obtain the mean and variance of the feature vector.

[0122] Randomly sample the mean and variance to obtain the feature distribution corresponding to the feature vector.

[0123] In one embodiment of the present application, Figure 7 As shown, the first determining module 603 is specifically configured to:

[0124] According to the pump valve state label as the input of the conditional variational autoencoder, the predicted state label is used as the output of the conditional variational autoencoder to generate an initial model of the conditional variational autoencoder.

[0125] Based on the difference between the pump valve state label and the predicted state label, as well as the difference between the feature distribution and the preset data distribution, the initial model is trained until the initial model meets the preset termination criteria to obtain the target model of the conditional variational autoencoder.

[0126] The target model parameters in the target model are used as the model parameters of the conditional variational autoencoder.

[0127] The present application proposes a pump valve fault diagnosis device, which obtains a full-cycle signal sample of the pump valve and its corresponding pump valve state label, and inputs the signal spectrum corresponding to the full-cycle signal sample into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum. The conditional variational autoencoder is trained according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and the model parameters of the conditional variational autoencoder are determined. The constructed decoder model is used as the pump valve fault diagnosis model, and the real-time signal spectrum and feature distribution corresponding to the real-time full-cycle signal sample of the pump valve are merged and input into the fault diagnosis model to obtain the real-time state prediction label of the pump valve and determine whether the pump valve is faulty. Thus, a fault diagnosis model is constructed based on the feature distribution of the signal spectrum to achieve accurate diagnosis of the pump valve state and improve the generalization performance of the fault diagnosis model.

[0128] According to an embodiment of the present application, the present application further provides a non-transitory computer-readable storage medium storing computer instructions, where the computer instructions are used to enable a computer to execute the pump valve fault diagnosis method in the embodiment of the present application.

[0129] The present application also proposes a computer program product, which implements the pump valve fault diagnosis method in the embodiment of the present application when the instruction processor in the computer program product is executed.

[0130] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0131] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0132] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A pump valve fault diagnosis method, characterized in that: The method comprises: Acquire a whole cycle signal sample of the pump valve and a pump valve status label corresponding to the whole cycle signal sample, and perform signal transformation on the whole cycle signal sample to obtain a signal spectrum corresponding to the whole cycle signal sample; Inputting the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum; Training the conditional variational autoencoder according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution to determine the model parameters of the conditional variational autoencoder; Based on the model parameters of the decoder in the conditional variational autoencoder, constructing the decoder model, and using the decoder model as the fault diagnosis model of the pump valve; The real-time signal spectrum corresponding to the real-time full-cycle signal sample of the pump valve and the characteristic distribution are combined and input into the fault diagnosis model to obtain a real-time state prediction label of the pump valve; Determine whether the pump valve is faulty based on the real-time status prediction tag of the pump valve.

2. The method according to claim 1, characterized in that Acquiring an entire cycle signal sample of the pump valve and a pump valve state label corresponding to the entire cycle signal sample, and performing a fast Fourier transform on the entire cycle signal sample to obtain a signal spectrum corresponding to the entire cycle signal sample, including: Acquire a vibration signal corresponding to the pump valve, and a key phase signal of a crankshaft corresponding to the pump valve; Performing windowing processing on the vibration signal and the key phase signal to obtain a full cycle signal sample of the pump valve; Marking the pump valve state corresponding to the full cycle signal sample to obtain a pump valve state label of the pump valve; Performing a fast Fourier transform on the integer-cycle signal samples to obtain an initial signal spectrum corresponding to the integer-cycle signal samples; The initial signal spectrum is normalized to obtain a normalized signal spectrum.

3. The method according to claim 1, characterized in that The conditional variational autoencoder includes an encoder, a latent feature space representation layer, and a decoder. The signal spectrum is input into the conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum, including: Inputting the signal spectrum into the encoder to obtain a feature vector corresponding to the signal spectrum; Inputting the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector; The feature distribution and the signal spectrum are input into the decoder to obtain a predicted state label corresponding to the signal spectrum.

4. The method according to claim 3, characterized in that Inputting the feature vector into the latent feature space representation layer to sample the feature distribution corresponding to the feature vector includes: Inputting the feature vector into the latent feature space representation layer to obtain the mean and variance of the feature vector; Random distribution sampling is performed on the mean and variance to obtain a feature distribution corresponding to the feature vector.

5. The method according to claim 1, wherein The step of training the conditional variational autoencoder based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution to determine the model parameters of the conditional variational autoencoder includes: Using the pump valve state label as the input of the conditional variational autoencoder and the predicted state label as the output of the conditional variational autoencoder to generate an initial model of the conditional variational autoencoder; Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the initial model is trained until the initial model meets a preset termination criterion to obtain a target model of the conditional variational autoencoder; The target model parameters in the target model are used as model parameters of the conditional variational autoencoder.

6. A pump valve fault diagnosis device, characterized in that: The device comprises: an acquisition module, configured to acquire a whole-cycle signal sample of the pump valve and a pump valve status label corresponding to the whole-cycle signal sample, and perform signal transformation on the whole-cycle signal sample to obtain a signal spectrum corresponding to the whole-cycle signal sample; A first generation module is used to input the signal spectrum into a conditional variational autoencoder to obtain a feature distribution and a predicted state label corresponding to the signal spectrum; a first determining module, configured to train the conditional variational autoencoder according to the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, and determine the model parameters of the conditional variational autoencoder; A construction module, configured to construct a decoder model based on model parameters of a decoder in the conditional variational autoencoder, and use the decoder model as a fault diagnosis model for the pump valve; a second generating module, configured to combine the real-time signal spectrum corresponding to the real-time full-cycle signal sample of the pump valve and the characteristic distribution and input the combined signals into the fault diagnosis model to obtain a real-time state prediction label of the pump valve; The second determination module is configured to determine whether the pump valve is faulty based on a real-time state prediction tag of the pump valve.

7. The device according to claim 6, characterized in that The acquisition module is specifically used to: Acquire a vibration signal corresponding to the pump valve, and a key phase signal of a crankshaft corresponding to the pump valve; Performing windowing processing on the vibration signal and the key phase signal to obtain a full cycle signal sample of the pump valve; Marking the pump valve state corresponding to the full cycle signal sample to obtain a pump valve state label of the pump valve; Performing a fast Fourier transform on the integer-cycle signal samples to obtain an initial signal spectrum corresponding to the integer-cycle signal samples; The initial signal spectrum is normalized to obtain a normalized signal spectrum.

8. The device according to claim 6, characterized in that The conditional variational autoencoder includes an encoder, a latent feature space representation layer and a decoder, and the first generation module includes: An input unit, configured to input the signal spectrum into the encoder to obtain a feature vector corresponding to the signal spectrum; a sampling unit, configured to input the feature vector into the potential feature space representation layer to sample a feature distribution corresponding to the feature vector; A generating unit is configured to input the feature distribution and the signal spectrum into the decoder to obtain a predicted state label corresponding to the signal spectrum.

9. The device according to claim 8, characterized in that The sampling unit is specifically used for: Inputting the feature vector into the latent feature space representation layer to obtain the mean and variance of the feature vector; Random distribution sampling is performed on the mean and variance to obtain a feature distribution corresponding to the feature vector.

10. The device according to claim 6, characterized in that The first determining module is specifically configured to: Using the pump valve state label as the input of the conditional variational autoencoder and the predicted state label as the output of the conditional variational autoencoder to generate an initial model of the conditional variational autoencoder; Based on the difference between the pump valve state label and the predicted state label and the difference between the feature distribution and the preset data distribution, the initial model is trained until the initial model meets a preset termination criterion to obtain a target model of the conditional variational autoencoder; The target model parameters in the target model are used as model parameters of the conditional variational autoencoder.

Images