Steam turbine set vibration fault diagnosis method and system

Abstract

The invention provides a steam turbine set vibration fault diagnosis method and system, and belongs to the technical field of thermal power plant fault diagnosis. The method comprises the following steps: inputting a to-be-identified vibration signal of a turboset into a trained turboset vibration fault diagnosis model for vibration fault diagnosis; the training rule of the steam turbine set vibration fault diagnosis model comprises the following steps: extracting a discrete wavelet transform coefficient from a steam turbine set vibration signal to obtain a DWTC image, and converting the steam turbine set vibration signal into a vibration signal image; taking the DWTC image and the vibration signal image as inputs of two branches of a double-flow convolutional neural network to perform high-level feature learning, and fusing high-level feature images output by the two branches of the double-flow convolutional neural network after preset processing; and inputting the fused image features obtained by fusion into a preset classifier, and outputting a corresponding turboset vibration fault type. The method and the system have the advantages of high accuracy, good generalization ability and strong robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of thermal power plants, and in particular to a method for diagnosing vibration faults of steam turbine units, a system for diagnosing vibration faults of steam turbine units, a machine-readable storage medium and an electronic device. Background Art

[0002] Steam turbine units are key production equipment in thermal power plants. They are complex in structure, have harsh working conditions (high temperature, high pressure, high speed), and have high requirements for continuous operation, making them prone to failure. In order to ensure the safe and stable operation of steam turbine units, it is of great practical significance to accurately and quickly diagnose the vibration failure of steam turbine units.

[0003] At present, the commonly used diagnostic methods for vibration faults of steam turbine units include expert systems, support vector machines (SVM) and neural networks, but these traditional intelligent diagnostic methods face problems such as large amount of sampling data, difficulty in extracting signal features and insufficient fault training samples. Therefore, how to accurately and quickly diagnose the operating status of steam turbine units, which have complex structures, harsh working conditions (high temperature, high pressure, high speed) and high continuous operation requirements, in order to improve the operation and maintenance efficiency of steam turbine units and accelerate the realization of fully intelligent operation of thermal power plants is an urgent problem to be solved. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method and system for diagnosing vibration faults of a steam turbine unit, so as to at least solve the above-mentioned problem of how to accurately and quickly diagnose the operating status of a steam turbine unit, which has the characteristics of complex structure, harsh working conditions and high continuous operation requirements, so as to improve the operation and maintenance efficiency of the steam turbine unit.

[0005] In order to achieve the above object, the present invention provides a first aspect of a steam turbine unit vibration fault diagnosis method, comprising:

[0006] Inputting the vibration signal to be identified of the steam turbine unit into the trained vibration fault diagnosis model of the steam turbine unit to perform vibration fault diagnosis, and obtaining a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the vibration fault type of the steam turbine unit;

[0007] The training rules of the turbine unit vibration fault diagnosis model include:

[0008] The vibration signals of the steam turbine units in the multiple sampling windows acquired in advance are used as the model input, and the vibration fault types of the steam turbine units corresponding to the vibration signals of the steam turbine units are used as the model output, and the steam turbine unit vibration fault diagnosis model is trained, including:

[0009] After extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, image conversion is performed to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images;

[0010] The DWTC image and the vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are respectively used as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and the high-level feature images output by the two branches of the two-stream convolutional neural network are subjected to preset processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features;

[0011] The fused image features are input into the preset classifier to identify the vibration fault category and output the corresponding steam turbine unit vibration fault type.

[0012] Optionally, the above-mentioned steam turbine unit vibration signal acquisition rule includes:

[0013] Obtaining a vibration signal of a steam turbine unit from a DCS of a thermal power plant and / or obtaining a vibration signal of a steam turbine unit corresponding to an experimental result through a vibration fault simulation experiment of the steam turbine unit;

[0014] All steam turbine unit vibration signals are combined through unified normalization processing.

[0015] Optionally, the training rules of the above steam turbine unit vibration fault diagnosis model also include:

[0016] Divide the combined steam turbine unit vibration signal into a training set and a test set;

[0017] The training set is used to train the steam turbine unit vibration fault diagnosis model, and the test set is used to test the trained steam turbine unit vibration fault diagnosis model.

[0018] Optionally, the test set includes steam turbine unit vibration signals in multiple sampling windows corresponding to various steam turbine unit vibration fault types and steam turbine unit vibration fault types corresponding to each steam turbine unit vibration signal;

[0019] The above test set is used to test the trained steam turbine unit vibration fault diagnosis model, including:

[0020] The steam turbine unit vibration signal in each sampling window is used as the input of the trained steam turbine unit vibration fault diagnosis model, and it is determined whether the steam turbine unit vibration fault type output by the trained steam turbine unit vibration fault diagnosis model is consistent with the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal;

[0021] The vibration fault type of the steam turbine unit includes at least one or more of a normal working state type, a bearing inner ring fault type, a bearing outer ring fault type and a rolling element fault type.

[0022] Optionally, the above discrete wavelet transform rules include:

[0023] Based on the Daubechies wavelet basis function, the vibration signal of the steam turbine unit in each sampling window is decomposed by a multi-scale one-dimensional discrete wavelet transform; wherein the scale of the one-dimensional discrete wavelet transform decomposition is set to k = log 2 N, where k is the scale of the one-dimensional discrete wavelet transform decomposition and N is the length of the sampling window;

[0024] For the vibration signal of the steam turbine unit in each sampling window, the high-frequency coefficients of the vibration signal of the steam turbine unit at the scale of k one-dimensional discrete wavelet transform decompositions and the low-frequency coefficients at the scale of the kth one-dimensional discrete wavelet transform decomposition are concatenated into a one-dimensional vector containing M elements as the discrete wavelet transform coefficients extracted from the vibration signal of the steam turbine unit.

[0025] Optionally, the above-mentioned preset processing includes feature flattening processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing include flattened high-level features corresponding to the DWTC image and flattened high-level features corresponding to the vibration signal image.

[0026] Optionally, the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features, including:

[0027] The high-level feature maps output by the two branches of the two-stream convolutional neural network after preset processing

[0028] y=H f (H v (x;θ v ),H w (W;θ w );θ f ) like according to the formula x∈R N×C ,W∈R M×C Perform fusion to obtain fused image features;

[0029] Where x represents the vibration signal of the steam turbine unit within a sampling window of length N, y is the fusion image feature of the final output, W is the DWTC image extracted from x, and H v is the branch of the two-stream convolutional neural network used to process vibration signal images, θ v H v Parameters, H wis the branch of the two-stream convolutional neural network used to process DWTC images, θ w H w Parameters, H f is a high-level feature fusion module for fusing the flattened high-level features corresponding to the DWTC image and the flattened high-level features corresponding to the vibration signal image, θ f H f Parameters.

[0030] A second aspect of the present invention provides a steam turbine unit vibration fault diagnosis system, comprising:

[0031] A vibration fault diagnosis module is used to input the vibration signal to be identified by the steam turbine unit into the trained steam turbine unit vibration fault diagnosis model to perform vibration fault diagnosis and obtain a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the type of vibration fault of the steam turbine unit;

[0032] The training rules of the turbine unit vibration fault diagnosis model include:

[0033] The vibration signals of the steam turbine units in the multiple sampling windows acquired in advance are used as the model input, and the vibration fault types of the steam turbine units corresponding to the vibration signals of the steam turbine units are used as the model output, and the steam turbine unit vibration fault diagnosis model is trained, including:

[0034] After extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, image conversion is performed to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images;

[0035] The DWTC image and the vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are respectively used as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and the high-level feature images output by the two branches of the two-stream convolutional neural network are subjected to preset processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features;

[0036] The fused image features are input into the preset classifier to identify the vibration fault category and output the corresponding steam turbine unit vibration fault type.

[0037] In a third aspect of the present invention, a machine-readable storage medium is provided, on which instructions are stored, and when the instructions are executed by a processor, the processor is configured to execute the above-mentioned steam turbine unit vibration fault diagnosis method.

[0038] In a fourth aspect of the present invention, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steam turbine unit vibration fault diagnosis method when executing the computer program.

[0039] Through the above technical scheme, a method and system for diagnosing vibration faults of a steam turbine unit is provided, which identifies the vibration signal to be identified of the steam turbine unit through a trained steam turbine unit vibration fault diagnosis model, so as to diagnose the vibration fault of the steam turbine unit and determine the type of vibration fault of the steam turbine unit. The training rules of the steam turbine unit vibration fault diagnosis model are as follows: the vibration signals of the steam turbine unit in multiple sampling windows acquired in advance are used as the model input, and the type of vibration fault of the steam turbine unit corresponding to each steam turbine unit vibration signal is used as the model output to train the vibration fault diagnosis model of the steam turbine unit. Among them, after the discrete wavelet transform coefficients are extracted from the vibration signals of the steam turbine unit in multiple sampling windows acquired in advance, the image conversion is performed to obtain the DWTC images corresponding to the vibration signals of the steam turbine unit in each sampling window, and the vibration signals of the steam turbine unit in each sampling window are converted into vibration signal images. The DWTC image and vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are used as the input of the two branches of the two-stream convolutional neural network for high-level feature learning, and the high-level feature images output by the two branches of the two-stream convolutional neural network are fused after preset processing to obtain fused image features. The fused image features are input into the preset classifier for vibration fault category recognition, and the corresponding steam turbine unit vibration fault type is output. Thus, the training of the steam turbine unit vibration fault diagnosis model is completed. In the vibration fault diagnosis process of the steam turbine unit, the method and system not only utilize the time-frequency domain information of the vibration signal provided by the discrete wavelet transform coefficients, but also utilize the high-level convolution features obtained by CNN from the vibration signal image corresponding to the original steam turbine unit vibration signal through feature learning, thereby realizing accurate and rapid operation status diagnosis for the steam turbine unit, which has the characteristics of complex structure, harsh working conditions (high temperature, high pressure, high speed) and high continuous operation requirements, so as to improve the operation and maintenance efficiency of the steam turbine unit and accelerate the realization of fully intelligent operation of thermal power plants. The method and system have the advantages of high accuracy, good generalization ability and strong robustness.

[0040] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0042] Figure 1 It is a flow chart of a steam turbine unit vibration fault diagnosis method provided by one embodiment of the present invention;

[0043] Figure 2 4 different steam turbine unit vibration signal waveforms provided by one embodiment of the present invention;

[0044] Figure 3 is a process diagram of converting a one-dimensional vibration signal into a two-dimensional grayscale image provided by an embodiment of the present invention;

[0045] Figure 4 It is a structural diagram of a two-dimensional image dual-stream convolutional neural network provided by one embodiment of the present invention;

[0046] Figure 5 It is a flow chart of a method for diagnosing vibration faults of a steam turbine unit based on a two-dimensional image double-stream convolutional neural network provided by an embodiment of the present invention;

[0047] Figure 6 It is a flow chart of another method for diagnosing vibration faults of steam turbine units based on two-dimensional image double-stream convolutional neural network provided by an embodiment of the present invention;

[0048] Figure 7 It is a schematic diagram of the structure of an electronic device provided by a preferred embodiment of the present invention.

[0049] Description of Reference Numerals

[0050] 10 - electronic device, 100 - processor, 101 - memory, 102 - computer program. DETAILED DESCRIPTION

[0051] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0052] Figure 1 FIG. 1 is a flow chart of a method for diagnosing vibration faults of a steam turbine unit provided by one embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for diagnosing vibration faults of a steam turbine unit, comprising:

[0053] S100: inputting the vibration signal to be identified of the steam turbine unit into the trained vibration fault diagnosis model of the steam turbine unit to perform vibration fault diagnosis, and obtain a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the vibration fault type of the steam turbine unit;

[0054] The training rules of the turbine unit vibration fault diagnosis model include:

[0055] S200: Using the pre-acquired steam turbine unit vibration signals in multiple sampling windows as model input, using the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal as model output, and training the steam turbine unit vibration fault diagnosis model, including:

[0056] S201: extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, performing image conversion to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and converting the steam turbine unit vibration signals in each sampling window into vibration signal images;

[0057] Specifically, discrete wavelet transform coefficients (DWTC) are extracted from the turbine vibration signal in each sampling window through discrete wavelet transform, and then converted into a DWTC image, and the original turbine vibration signal in each sampling window (i.e., turbine vibration signal) is converted into a vibration signal image. Wherein, the length of each sample data (i.e., the length of the sampling window of each turbine vibration signal) is 1024, and the one-dimensional time series signal will be converted into a 32×32 image.

[0058] In some implementations of this embodiment, the above discrete wavelet transform rules include: based on the Daubechies wavelet basis function, performing multi-scale one-dimensional discrete wavelet transform decomposition on the vibration signal of the steam turbine unit in each sampling window; wherein the scale of the one-dimensional discrete wavelet transform decomposition is set to k=log 2 N, wherein k is the scale of one-dimensional discrete wavelet transform decomposition, and N is the length of the sampling window; for the vibration signal of the steam turbine unit in each sampling window, the high-frequency coefficients of the vibration signal of the steam turbine unit at the scales of k one-dimensional discrete wavelet transform decompositions and the low-frequency coefficients at the scale of the kth one-dimensional discrete wavelet transform decomposition are concatenated into a one-dimensional vector containing M elements as the discrete wavelet transform coefficients extracted from the vibration signal of the steam turbine unit.

[0059] S202: using the DWTC image and the vibration signal image corresponding to the steam turbine unit vibration signal in each sampling window as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and fusing the high-level feature images output by the two branches of the two-stream convolutional neural network after preset processing to obtain fused image features;

[0060] Specifically, the vibration signal image stream (i.e., a set of multiple vibration signal images) and the DWTC image stream (i.e., a set of multiple DWTC images) are used as the input of two branches of a two-stream convolutional neural network (CNN) for high-level feature learning. The CNN branch structure for processing the vibration signal image stream includes 2 convolutional layers and 2 local connection layers. Each convolutional layer uses a 3×3 two-dimensional filter. The number of convolutional feature maps generated by each convolutional layer is 32. After the two convolutional layers, there are 2 local connection layers using 1×1 filters. The number of feature maps generated by each local connection layer is 32. The GengNet structure is the most widely used CNN structure here. The CNN branch used to process the DWTC image stream consists of 2 convolutional layers and 2 local connection layers. Each convolutional layer uses a 2×2 two-dimensional filter (convolution kernel) to convolve the DWTC image, and the number of convolution feature maps generated by each convolutional layer is 32. The two convolutional layers are followed by two local connection layers using 1×1 filters, and the number of feature maps generated by each local connection layer is 32. For each convolutional layer and local connection layer of the two CNN branches, batch normalization (BatchNormalization, BN) is widely used, and the ReLU (Rectified Linear Unit) nonlinear activation function is set. The main purpose of batch normalization is to standardize the input data samples of each batch to avoid the problem of data distribution changes caused by changes in parameters of each layer in CNN. The ReLU nonlinear activation function mainly avoids the gradient vanishing problem and accelerates the convergence speed by nonlinearly transforming the neuron input. In addition, random dropout regularization is set for the second local connection layer of each branch to avoid overfitting. Figure 4 As shown, Figure 4This is a structural diagram of a two-dimensional image dual-stream convolutional neural network provided by an embodiment of the present invention. The vibration signal image stream and the DWTC image stream are respectively used as the input of the two branches of the dual-stream convolutional neural network (CNN) to learn high-level features. The feature vectors output by the two branches are spliced ​​and input into a fully connected layer with the number of neurons equal to the number of vibration fault categories to be identified. The output of the fully connected layer is used as the input of the Softmax classifier to identify the vibration fault category.

[0061] In some implementations of this embodiment, the above-mentioned preset processing includes feature flattening processing, and the high-level feature images output by the two branches of the dual-stream convolutional neural network after the preset processing include flattened high-level features corresponding to the DWTC image and flattened high-level features corresponding to the vibration signal image.

[0062] Specifically, the vibration signal image stream (i.e., a set containing multiple vibration signal images) and the DWTC image stream (i.e., a set containing multiple DWTC images) are respectively used as the input of the two branches of the two-stream convolutional neural network (CNN) for high-level feature learning. The high-level feature images output by the two CNN branches are flattened and then input into the high-level feature fusion module for fusion. The outputs of the two CNN branches are fused through the high-level feature fusion module. The feature vectors output by the two branches (i.e., the high-level feature images output by the two branches) are spliced ​​and input into a fully connected layer whose number of neurons is equal to the number of vibration fault categories to be identified. The output of the fully connected layer is used as the input of the Softmax classifier for vibration fault category identification. In the above-mentioned high-level feature fusion module, the high-level feature images learned by the two branches of the two-stream convolutional neural network from the vibration signal image stream and the DWTC image stream are fused twice in the feature layer, thereby realizing more sufficient and frequent information interaction and transmission between the two data streams (i.e., the vibration signal image stream and the DWTC image stream).

[0063] In some implementations of this embodiment, the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain the fused image features, including: the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused according to

[0064] y=H f (H v (x;θ v ),H w (W;θ w );θ f )

[0065] According to the formula x∈R N×C ,W∈R M×C Fusion is performed to obtain fused image features; where x represents the vibration signal of the steam turbine unit within a sampling window of length N, y is the final output fused image feature, W is the DWTC image extracted from x, and H v is the branch of the two-stream convolutional neural network used to process vibration signal images, θ v H v Parameters, H w is the branch of the two-stream convolutional neural network used to process DWTC images, θ w H w Parameters, H f is a high-level feature fusion module for fusing the flattened high-level features corresponding to the DWTC image and the flattened high-level features corresponding to the vibration signal image, θ f H f Parameters. Among them, H v and H w Keep parameters independent of each other during training.

[0066] Specifically, the outputs of the two CNN branches are fused through a high-level feature fusion module. The high-level feature fusion module contains two CNN branches, each of which is composed of two fully connected layers. Among them, the number of neurons in the two fully connected layers of the first branch is 512 and 128 respectively, and the input of this branch is the flattened high-level features output by the CNN branch corresponding to the vibration signal image stream; the number of neurons in the two fully connected layers of the second branch is 1024 and 512 respectively, and the input of this branch is a feature vector formed by concatenating the flattened high-level features output by the CNN branch corresponding to the vibration signal image stream and the flattened high-level features output by the CNN branch corresponding to the DWTC image stream.

[0067] S203: Input the fused image features into a preset classifier to identify the vibration fault category, and output the corresponding steam turbine unit vibration fault type.

[0068] Specifically, the method identifies the vibration signal to be identified by the steam turbine unit through the trained steam turbine unit vibration fault diagnosis model, so as to perform vibration fault diagnosis on the steam turbine unit and determine the type of vibration fault of the steam turbine unit. The training rules of the steam turbine unit vibration fault diagnosis model are as follows: the steam turbine unit vibration signals in multiple sampling windows acquired in advance are used as model input, and the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal is used as model output to train the steam turbine unit vibration fault diagnosis model. Among them, after the discrete wavelet transform coefficients are extracted from the steam turbine unit vibration signals in multiple sampling windows acquired in advance, image conversion is performed to obtain the DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images. The DWTC image and vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are used as the input of the two branches of the two-stream convolutional neural network for high-level feature learning, and the high-level feature images output by the two branches of the two-stream convolutional neural network are fused after preset processing to obtain fused image features. The fused image features are input into the preset classifier for vibration fault category recognition, and the corresponding steam turbine unit vibration fault type is output. Thus, the training of the steam turbine unit vibration fault diagnosis model is completed. In the vibration fault diagnosis process of steam turbine units, this method not only utilizes the time-frequency domain information of the vibration signal provided by the discrete wavelet transform coefficients, but also utilizes the high-level convolution features obtained by CNN from the vibration signal image corresponding to the original steam turbine unit vibration signal through feature learning. It realizes accurate and rapid operation status diagnosis for steam turbine units, which have complex structures, harsh working conditions (high temperature, high pressure, high speed) and high continuous operation requirements, so as to improve the operation and maintenance efficiency of steam turbine units and accelerate the realization of fully intelligent operation of thermal power plants. It has the advantages of high accuracy, good generalization ability and strong robustness.

[0069] In the above implementation process, the embodiment of the present application provides a method for diagnosing vibration faults of a steam turbine unit. The method identifies the vibration signal to be identified of the steam turbine unit through a trained steam turbine unit vibration fault diagnosis model. The two convolutional neural networks (Convolutional Neural Networks) of the steam turbine unit vibration fault diagnosis model are A convolutional neural network (CNN) model branch learns high-level features from the image converted from the original vibration signal and the image converted from the discrete wavelet transform coefficient, respectively, and fuses the high-level features learned by the two branches through a high-level feature fusion module. The fused high-level features are input into a classifier for vibration fault diagnosis, which includes the following steps: 1) taking each vibration fault type of a steam turbine unit and a parameter set when the fault occurs as training samples, wherein the training samples include multiple fault samples corresponding to multiple fault types, and each fault sample contains multiple steam turbine unit parameters; 2) constructing a steam turbine unit vibration fault diagnosis model based on a two-dimensional image dual-stream convolutional neural network, and using the training samples for training; 3) extracting discrete wavelet transform coefficients from each sample steam turbine unit vibration signal, and converting them into two-dimensional grayscale images respectively with the original steam turbine unit vibration signal; 4) the two image streams are used as the input of the two branches of the dual-stream convolutional neural network for high-level feature learning; 5) fusing the high-level features learned by the two branches through a high-level feature fusion module, and realizing the vibration fault type diagnosis of the steam turbine unit according to the fused features. In the process of vibration fault diagnosis, the time-frequency domain information of the vibration signal provided by the discrete wavelet transform coefficients is used, and the high-level convolution features obtained from the original feature learning by CNN are also used. Compared with the existing technology, this method has the advantages of high accuracy, good generalization ability, and strong robustness. This method converts the original steam turbine unit vibration signal into a two-dimensional grayscale image for processing, fully considering the time-frequency information contained in the steam turbine unit vibration signal, thereby improving the authenticity of feature extraction. The steam turbine unit vibration signal and the signal discrete wavelet transform coefficients containing the signal time-frequency domain information are converted into images, respectively input into the two branches of the dual-stream CNN model to learn high-level features, and through the high-level feature fusion module, more sufficient and frequent information interaction and transmission between the high-level features of the two data streams are realized. The dual-stream CNN model, the vibration signal image stream model of the single-stream CNN, and the DWTC image stream model of the single-stream CNN are compared and tested. This method can accurately complete the fault diagnosis of various vibration types of steam turbine units, and has a higher fault recognition rate than other methods.

[0070] Please refer to Figure 3 , Figure 3The process of converting a one-dimensional vibration signal into a two-dimensional grayscale image is as follows: the first M signal points in the one-dimensional time domain signal are arranged in a row as the first row of grayscale image pixels, and the last M signal points are arranged in a row as the second row of grayscale image pixels, and so on, until the M×(M-1)th to Mth row of grayscale image pixels are taken. 2 signal points as the Mth column of grayscale image pixels, and the conversion of the first grayscale image is completed. Its expression is as follows: Where P(j,k) represents the pixel intensity of the image, K = 1, 2, ... M, L(i) represents the value of a segment of the signal, i = 1, 2, ... M × M, and round(·) represents the rounding function, which normalizes the pixel intensity of the grayscale image from 0 to 255, i.e., the image of the grayscale image. The wavelet transform scale k is set to k = log 2 N, where N is the sampling window length, the high-frequency coefficients of each pass at all k wavelet transform scales and the low-frequency coefficients at the kth wavelet transform scale are concatenated into a one-dimensional vector containing M elements as the wavelet transform coefficient vector extracted from the vibration signal, and then converted into a DWTC image. The advantage of this method is that it can effectively explore its two-dimensional features from a one-dimensional signal, can be calculated without any predefined parameters, and minimize the influence of expert experience.

[0071] It should be noted that with the increase in the sample size of turbine unit vibration fault data sets and the development of deep machine learning technology, the latest research in this field has begun to focus on feature learning methods represented by deep neural networks, which can automatically learn high-level abstract features from a large number of input data samples through multiple nonlinear transformations. Therefore, it can get rid of the manual extraction and selection process of time domain, frequency domain or time-frequency domain signal features to a certain extent, and realize "end-to-end" fault diagnosis.

[0072] In some implementations of this embodiment, the above-mentioned rules for obtaining the turbine unit vibration signal include: obtaining the turbine unit vibration signal from the DCS of the thermal power plant and / or obtaining the turbine unit vibration signal of the corresponding experimental result through the turbine unit vibration fault simulation experiment; merging all turbine unit vibration signals through unified normalization processing.

[0073] In some implementations of the present embodiment, the training rules of the above-mentioned steam turbine unit vibration fault diagnosis model also include: dividing the merged steam turbine unit vibration signal into a training set and a test set; using the training set to train the steam turbine unit vibration fault diagnosis model, and using the test set to test the trained steam turbine unit vibration fault diagnosis model.

[0074] Specifically, two methods are used to obtain the vibration signal of the steam turbine unit. One is to obtain it from the distributed control system (DCS) of the thermal power plant, and the other is to collect the vibration signal of the steam turbine unit through the data acquisition software in the laboratory simulation of the vibration fault experiment of the steam turbine unit to obtain the vibration signal of the steam turbine unit. The vibration signal of the steam turbine unit mainly includes: the vibration signal of the steam turbine unit in the case of bearing inner ring fault, bearing outer ring fault and rolling element fault, and the vibration signal of the steam turbine unit in the normal state (please refer to Figure 2 , Figure 2 It is a waveform diagram of four different steam turbine unit vibration signals provided by an embodiment of the present invention. The vibration of the steam turbine unit mainly comes from the change of the bearing state of the unit, mainly including the vibration signal of the steam turbine unit in the case of bearing inner ring fault, outer ring fault, rolling element fault, etc., as well as the vibration signal under normal state). Then, the vibration signals of the steam turbine unit obtained by these two methods are merged through unified normalization processing (specifically, the vibration signals of the steam turbine unit obtained by these two methods belong to heterogeneous data merging, so the two types of data are merged through unified normalization processing), and the merged steam turbine unit vibration signals are randomly divided into training set and test set according to a certain ratio (for example, the ratio between training set and test set is 8:2). Among them, the steam turbine unit vibration fault diagnosis model is constructed based on a two-dimensional image dual-stream convolutional neural network, and the steam turbine unit vibration fault diagnosis model is trained using the training set, and the trained steam turbine unit vibration fault diagnosis model is used for subsequent vibration fault diagnosis.

[0075] In some implementations of the present embodiment, the test set includes turbine unit vibration signals in multiple sampling windows corresponding to various turbine unit vibration fault types and turbine unit vibration fault types corresponding to each turbine unit vibration signal; the test set is used to test the trained turbine unit vibration fault diagnosis model, including: using the turbine unit vibration signal in each sampling window as the input of the trained turbine unit vibration fault diagnosis model, and judging whether the turbine unit vibration fault type output by the trained turbine unit vibration fault diagnosis model is consistent with the turbine unit vibration fault type corresponding to each turbine unit vibration signal; wherein the turbine unit vibration fault type includes at least one or more of a normal working state type, a bearing inner ring fault type, a bearing outer ring fault type and a rolling element fault type.

[0076] Specifically, the process of testing the trained steam turbine unit vibration fault diagnosis model is as follows: the steam turbine unit vibration signal in the test set is used as the input of the trained two-stream convolutional neural network, and the output is the steam turbine unit vibration fault type, including normal working state and bearing inner ring fault, outer ring fault, rolling element fault and other fault types. Compare whether the steam turbine unit vibration fault type output by the trained steam turbine unit vibration fault diagnosis model is consistent with the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal, so as to test the accuracy of the trained steam turbine unit vibration fault diagnosis model, and adjust the model parameters of the trained steam turbine unit vibration fault diagnosis model according to the test results, so as to further improve the recognition accuracy of the trained steam turbine unit vibration fault diagnosis model.

[0077] Please refer to Figure 5 and Figure 6 , Figure 5 It is a flow chart of a method for diagnosing vibration faults of a steam turbine unit based on a two-dimensional image double-stream convolutional neural network provided by an embodiment of the present invention. Figure 6 This is a flow chart of another method for diagnosing vibration faults of steam turbine units based on a two-dimensional image double-stream convolutional neural network provided by an embodiment of the present invention. This application includes the following steps:

[0078] 1) Acquisition of vibration fault data of steam turbine unit: Two methods are used to acquire data, one is to acquire data from the distributed control system (DCS) of the thermal power plant, and the other is to acquire data from the vibration signal of the steam turbine unit through data acquisition software in the laboratory simulation of the vibration fault experiment of the steam turbine unit. The data mainly include: vibration signals of the steam turbine unit in cases of bearing inner ring fault, outer ring fault, rolling element fault, etc., as well as vibration signals in normal state. Then the two types of data are merged through unified normalization processing, and the data are randomly divided into training set and test set according to a certain proportion. In the present invention, a total of 1,320 pieces of steam turbine unit vibration signal data are collected for experimental verification, which are divided into training set and test set, with 792 and 528 samples respectively. Among them, the training set contains 177 normal state data and 615 fault state data. The fault types are mainly divided into three types: bearing inner ring fault, outer ring fault, and rolling element fault, and the fault state data are subdivided into 237 pieces of outer ring fault data, 135 pieces of outer ring fault data, and 243 pieces of rolling element fault data;

[0079] 2) Extract discrete wavelet transform coefficients (DWTC) from the vibration signal of the steam turbine unit in each sampling window through discrete wavelet transform, and then convert it into a DWTC image, and at the same time convert the original steam turbine unit vibration signal in each sampling window into a vibration signal image. In the present invention, the sample data length is 1024, and the one-dimensional time series signal will be converted into a 32×32 image;

[0080] 3) The vibration signal image stream and the DWTC image stream are used as the input of the two branches of the two-stream convolutional neural network (CNN) for high-level feature learning. The CNN branch structure for processing the vibration signal image stream includes 2 convolutional layers and 2 local connection layers. Each convolutional layer uses a 3×3 two-dimensional filter. The number of convolutional feature maps generated by each convolutional layer is 32. The two convolutional layers are followed by 2 local connection layers using 1×1 filters. The number of feature maps generated by each local connection layer is 32. The CNN branch for processing the DWTC image stream consists of 2 convolutional layers and 2 local connection layers. Each convolutional layer uses a 2×2 two-dimensional filter (convolution kernel) to convolve the DWTC image. The number of convolutional feature maps generated by each convolutional layer is 32. The two convolutional layers are followed by 2 local connection layers using 1×1 filters. The number of feature maps generated by each local connection layer is 32. The high-level feature images output by the two CNN branches are flattened and then input into the high-level feature fusion module for fusion.

[0081] 4) The outputs of the two CNN branches are fused through the high-level feature fusion module. The high-level feature fusion module contains two CNN branches, each of which is composed of two fully connected layers. Among them: the number of neurons in the two fully connected layers of the first branch are 512 and 128 respectively, and the input of this branch is the flattened high-level features output by the CNN branch corresponding to the vibration signal image stream; the number of neurons in the two fully connected layers of the second branch are 1024 and 512 respectively, and the input of this branch is the feature vector formed by the concatenation of the flattened high-level features output by the CNN branch corresponding to the vibration signal image stream and the flattened high-level features output by the CNN branch corresponding to the DWTC image stream. Let x represent the vibration signal fusion in a sampling window of length N. The above process can be described by the following formula: Where: y is the high-level feature finally output by the CNN model; W is the DWTC image extracted from x; H v is the CNN branch used to process the original vibration signal image stream; θ v H v Parameters; H wis the CNN branch used to process the DWTC image stream; θ w H w Parameters; H f is a high-level feature fusion module used to fuse the output features of two CNN branches; θ f H f Parameters. v and H w Keep the parameters independent of each other during the training process. The feature vectors output by the two branches are concatenated and input into a fully connected layer with the number of neurons equal to the number of vibration fault categories to be identified. The output of the fully connected layer is used as the input of the Softmax classifier to identify the vibration fault category.

[0082] 5) In the above-mentioned high-level feature fusion module, the high-level features learned by the two branches of the dual-stream CNN from the vibration signal image stream and the DWTC image stream are fused twice, thereby achieving more sufficient and frequent information interaction and transmission between the two data streams;

[0083] 6) The feature vector set in the test sample set is used as the input of the dual-stream CNN classifier to optimize the neural network hyperparameter settings, and the output value of the Softmax classifier decision function is used to determine the vibration fault type of the steam turbine unit, including normal working conditions, and fault types such as bearing inner race fault, outer race fault bearing, and rolling element fault bearing;

[0084] 7) Stochastic Gradient Descent (SGD) was used to train the neural network, with the batch size set to 1000 and the number of iterations set to 28. In order to accelerate convergence, a learning rate decay strategy was adopted, that is, the initial learning rate was set to 0.1, and the learning rate was divided by 10 at the 16th iteration and the 24th iteration, respectively. For all hidden layers of the neural network using random inactivation regularization, the proportion of randomly inactivated neurons was 65%. Vibration signal images (based on single-stream CNN), DWTC images (based on single-stream CNN), and vibration signal images and DWTC images (based on dual-stream CNN) were used for comparative tests, and the accuracy rates were 91.4%, 93.8%, and 94.9%, respectively. The method of the present invention has a higher fault recognition rate.

[0085] The embodiment of the present invention provides a steam turbine unit vibration fault diagnosis system, comprising:

[0086] A vibration fault diagnosis module is used to input the vibration signal to be identified by the steam turbine unit into the trained steam turbine unit vibration fault diagnosis model to perform vibration fault diagnosis and obtain a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the type of vibration fault of the steam turbine unit;

[0087] The training rules of the turbine unit vibration fault diagnosis model include:

[0088] The vibration signals of the steam turbine units in the multiple sampling windows acquired in advance are used as the model input, and the vibration fault types of the steam turbine units corresponding to the vibration signals of the steam turbine units are used as the model output, and the steam turbine unit vibration fault diagnosis model is trained, including:

[0089] After extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, image conversion is performed to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images;

[0090] The DWTC image and the vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are respectively used as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and the high-level feature images output by the two branches of the two-stream convolutional neural network are subjected to preset processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features;

[0091] The fused image features are input into the preset classifier to identify the vibration fault category and output the corresponding steam turbine unit vibration fault type.

[0092] Specifically, the system identifies the vibration signal to be identified by the steam turbine unit through the trained steam turbine unit vibration fault diagnosis model, so as to perform vibration fault diagnosis on the steam turbine unit and determine the type of vibration fault of the steam turbine unit. The training rules of the steam turbine unit vibration fault diagnosis model are as follows: the steam turbine unit vibration signals in multiple sampling windows acquired in advance are used as model input, and the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal is used as model output to train the steam turbine unit vibration fault diagnosis model. Among them, after the discrete wavelet transform coefficients are extracted from the steam turbine unit vibration signals in multiple sampling windows acquired in advance, image conversion is performed to obtain the DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images. The DWTC image and vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are used as the input of the two branches of the two-stream convolutional neural network for high-level feature learning, and the high-level feature images output by the two branches of the two-stream convolutional neural network are fused after preset processing to obtain fused image features. The fused image features are input into the preset classifier for vibration fault category recognition, and the corresponding steam turbine unit vibration fault type is output. Thus, the training of the steam turbine unit vibration fault diagnosis model is completed. In the process of vibration fault diagnosis of steam turbine units, the system not only utilizes the time-frequency domain information of vibration signals provided by discrete wavelet transform coefficients, but also utilizes the high-level convolution features obtained by CNN from the vibration signal image corresponding to the original steam turbine unit vibration signal through feature learning. It realizes accurate and rapid operation status diagnosis for steam turbine units, which have complex structures, harsh working conditions (high temperature, high pressure, high speed) and high continuous operation requirements, so as to improve the operation and maintenance efficiency of steam turbine units and accelerate the realization of fully intelligent operation of thermal power plants. It has the advantages of high accuracy, good generalization ability and strong robustness.

[0093] The embodiment of the present invention further provides a machine-readable storage medium, on which instructions are stored. When the instructions are executed by the processor 100, the processor 100 is configured to execute the above-mentioned steam turbine unit vibration fault diagnosis method.

[0094] Machine-readable storage media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0095] The embodiment of the present invention further provides an electronic device 10, which includes a memory 101, a processor 100, and a computer program 102 stored in the memory 101 and executable on the processor 100. When the processor 100 executes the computer program 102, the above-mentioned steam turbine unit vibration fault diagnosis method is implemented.

[0096] like Figure 7 FIG. 1 is a schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 7 As shown, the electronic device 10 of this embodiment includes: a processor 100, a memory 101, and a computer program 102 stored in the memory 101 and executable on the processor 100. When the processor 100 executes the computer program 102, the steps in the above method embodiment are implemented. Alternatively, when the processor 100 executes the computer program 102, the functions of each module / unit in the above device embodiment are implemented.

[0097] Exemplarily, the computer program 102 may be divided into one or more modules / units, one or more modules / units are stored in the memory 101, and are executed by the processor 100 to complete the present invention. One or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program 102 in the electronic device 10. For example, the computer program 102 may be divided into a vibration fault diagnosis module.

[0098] The electronic device 10 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device 10 may include, but is not limited to, a processor 100 and a memory 101. Those skilled in the art will appreciate that Figure 7It is only an example of the electronic device 10 and does not constitute a limitation of the electronic device 10. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0099] The processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0100] The memory 101 may be an internal storage unit of the electronic device 10, such as a hard disk or memory of the electronic device 10. The memory 101 may also be an external storage device of the electronic device 10, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 10. Further, the memory 101 may also include both an internal storage unit of the electronic device 10 and an external storage device. The memory 101 is used to store computer programs and other programs and data required by the electronic device 10. The memory 101 may also be used to temporarily store data that has been output or is to be output.

[0101] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0102] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program 102 products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program 102 product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program 102 products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by the computer program 102 instructions. These computer program 102 instructions can be provided to a processor 100 of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor 100 of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program 102 instructions may also be stored in a computer readable memory 101 that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory 101 produce an article of manufacture including an instruction device that implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program 102 instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0107] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for diagnosing vibration faults of a steam turbine unit, characterized in that: include: Inputting the vibration signal to be identified of the steam turbine unit into the trained vibration fault diagnosis model of the steam turbine unit to perform vibration fault diagnosis, and obtaining a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the vibration fault type of the steam turbine unit; The training rules of the steam turbine unit vibration fault diagnosis model include: The vibration signals of the steam turbine units in the multiple sampling windows acquired in advance are used as the model input, and the vibration fault types of the steam turbine units corresponding to the vibration signals of the steam turbine units are used as the model output, and the steam turbine unit vibration fault diagnosis model is trained, including: After extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, image conversion is performed to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images; The DWTC image and the vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are respectively used as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and the high-level feature images output by the two branches of the two-stream convolutional neural network are subjected to preset processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features; The fused image features are input into the preset classifier to identify the vibration fault category and output the corresponding steam turbine unit vibration fault type.

2. The method for diagnosing vibration faults of a steam turbine unit according to claim 1, characterized in that: The steam turbine unit vibration signal acquisition rules include: Obtaining a vibration signal of a steam turbine unit from a DCS of a thermal power plant and / or obtaining a vibration signal of a steam turbine unit corresponding to an experimental result through a vibration fault simulation experiment of the steam turbine unit; All steam turbine unit vibration signals are combined through unified normalization processing.

3. The method for diagnosing vibration faults of a steam turbine unit according to claim 2, characterized in that: The training rules of the steam turbine unit vibration fault diagnosis model also include: Divide the combined steam turbine unit vibration signal into a training set and a test set; The training set is used to train the steam turbine unit vibration fault diagnosis model, and the test set is used to test the trained steam turbine unit vibration fault diagnosis model.

4. The method for diagnosing vibration faults of a steam turbine unit according to claim 3, characterized in that: The test set includes steam turbine unit vibration signals in a plurality of sampling windows corresponding to various steam turbine unit vibration fault types and steam turbine unit vibration fault types corresponding to each steam turbine unit vibration signal; The method of testing the trained steam turbine unit vibration fault diagnosis model using the test set includes: The steam turbine unit vibration signal in each sampling window is used as the input of the trained steam turbine unit vibration fault diagnosis model, and it is determined whether the steam turbine unit vibration fault type output by the trained steam turbine unit vibration fault diagnosis model is consistent with the steam turbine unit vibration fault type corresponding to each steam turbine unit vibration signal; The turbine unit vibration fault type includes at least one or more of a normal working state type, a bearing inner ring fault type, a bearing outer ring fault type and a rolling element fault type.

5. The method for diagnosing vibration faults of a steam turbine unit according to claim 1, characterized in that: The rules of the discrete wavelet transform include: Based on the Daubechies wavelet basis function, the vibration signal of the steam turbine unit in each sampling window is decomposed by a multi-scale one-dimensional discrete wavelet transform; wherein the scale of the one-dimensional discrete wavelet transform decomposition is set to k=log2N, wherein k is the scale of the one-dimensional discrete wavelet transform decomposition, and N is the length of the sampling window; For the vibration signal of the steam turbine unit in each sampling window, the high-frequency coefficients of the vibration signal of the steam turbine unit at the scale of k one-dimensional discrete wavelet transform decompositions and the low-frequency coefficients at the scale of the kth one-dimensional discrete wavelet transform decomposition are concatenated into a one-dimensional vector containing M elements as the discrete wavelet transform coefficients extracted from the vibration signal of the steam turbine unit.

6. The method for diagnosing vibration faults of a steam turbine unit according to claim 1, characterized in that: The preset processing includes feature flattening processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing include flattened high-level features corresponding to the DWTC image and flattened high-level features corresponding to the vibration signal image.

7. The method for diagnosing vibration faults of a steam turbine unit according to claim 6, characterized in that: The high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features, including: The high-level feature maps output by the two branches of the two-stream convolutional neural network after preset processing y=H f (H v (x;θ v ),H w (W;θ w );θ f ) like according to the formula x∈R N×C ,W∈R M×C Perform fusion to obtain fused image features; Where x represents the vibration signal of the steam turbine unit within a sampling window of length N, y is the fusion image feature of the final output, W is the DWTC image extracted from x, and H v is the branch of the two-stream convolutional neural network used to process vibration signal images, θ v H v Parameters, H w is the branch of the two-stream convolutional neural network used to process DWTC images, θ w H w Parameters, H f is a high-level feature fusion module for fusing the flattened high-level features corresponding to the DWTC image and the flattened high-level features corresponding to the vibration signal image, θ f H f Parameters.

8. A steam turbine unit vibration fault diagnosis system, characterized in that: include: A vibration fault diagnosis module, used for inputting the vibration signal to be identified by the steam turbine unit into the trained vibration fault diagnosis model of the steam turbine unit to perform vibration fault diagnosis and obtain a vibration fault diagnosis result; wherein the vibration fault diagnosis result includes the vibration fault type of the steam turbine unit; The training rules of the steam turbine unit vibration fault diagnosis model include: The vibration signals of the steam turbine units in the multiple sampling windows acquired in advance are used as the model input, and the vibration fault types of the steam turbine units corresponding to the vibration signals of the steam turbine units are used as the model output, and the steam turbine unit vibration fault diagnosis model is trained, including: After extracting discrete wavelet transform coefficients from the pre-acquired steam turbine unit vibration signals in multiple sampling windows through discrete wavelet transform, image conversion is performed to obtain DWTC images corresponding to the steam turbine unit vibration signals in each sampling window, and the steam turbine unit vibration signals in each sampling window are converted into vibration signal images; The DWTC image and the vibration signal image corresponding to the vibration signal of the steam turbine unit in each sampling window are respectively used as the input of the two branches of the two-stream convolutional neural network to learn high-level features, and the high-level feature images output by the two branches of the two-stream convolutional neural network are subjected to preset processing, and the high-level feature images output by the two branches of the two-stream convolutional neural network after the preset processing are fused to obtain fused image features; The fused image features are input into the preset classifier to identify the vibration fault category and output the corresponding steam turbine unit vibration fault type.

9. 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 steam turbine unit vibration fault diagnosis method according to any one of claims 1 to 7.

10. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for diagnosing vibration faults of a steam turbine unit as described in any one of claims 1 to 7 is implemented.

Images