Fault diagnosis method and device for steam turbine rotor, electronic equipment and storage medium

Abstract

The invention relates to a fault diagnosis method and device for a steam turbine rotor, electronic equipment and a storage medium, and the method comprises the steps: obtaining historical vibration data of the steam turbine rotor, and constructing a training sample set and a test sample set based on the historical vibration data and an actual state tag corresponding to the historical vibration data; training an initial fault diagnosis model constructed by a graph convolutional neural network by using the training sample set to obtain a trained initial fault diagnosis model; and testing the trained initial fault diagnosis model by using the test sample set, generating a final fault diagnosis model of the steam turbine rotor meeting a preset test condition, and diagnosing a fault result of the target steam turbine rotor by using the final fault diagnosis model. Therefore, the technical problems that in the related technology, the accuracy of fault diagnosis only depending on similarity measurement is limited, the complex relation between the fault type characteristics and the vibration signals is difficult to mine deeply through a model algorithm, and the error of the diagnosis result is large are solved.

Description

Technical Field

[0001] This application relates to the technical field of mechanical fault diagnosis in thermal power plants, and particularly relates to a fault diagnosis method, device, electronic device, and storage medium for a steam turbine rotor. Background Art

[0002] In related technologies, the fault diagnosis methods for steam turbine rotors can be divided into two categories: distance-based methods and model-based methods.

[0003] Among them, the distance-based method can perform fault diagnosis by calculating the similarity metric between the sample to be classified or verified and the historical samples, but the accuracy of fault diagnosis is limited.

[0004] For model-based methods, intelligent algorithms such as ANN (artificial neural network), SVM (support vector machine), etc. have limited feature extraction capabilities. Deep learning algorithms such as CNN (convolutional neural network), DBN (deep belief network), and CapsNets (capsule networks) are difficult to deeply mine the complex relationship between the fault type and the vibration signal, thus affecting the accuracy of fault diagnosis.

[0005] In summary, in related technologies, relying solely on similarity metrics for fault diagnosis has limited accuracy, and model algorithms are difficult to deeply mine the complex relationship between fault type features and vibration signals, resulting in a large error in the diagnosis result, which needs to be improved. Summary of the Invention

[0006] This application provides a fault diagnosis method, device, electronic device, and storage medium for a steam turbine rotor to solve the technical problem that in related technologies, relying solely on similarity metrics for fault diagnosis has limited accuracy, and model algorithms are difficult to deeply mine the complex relationship between fault type features and vibration signals, resulting in a large error in the diagnosis result.

[0007] The first aspect of the present application provides a fault diagnosis method for a steam turbine rotor, which is applied to the model construction stage. The method includes the following steps: obtaining historical vibration data of the steam turbine rotor, and constructing a training sample set and a test sample set based on the historical vibration data and the actual state labels corresponding to the historical vibration data; training an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set to obtain a trained initial fault diagnosis model; testing the trained initial fault diagnosis model using the test sample set to generate a final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to diagnose the fault result of the target steam turbine rotor using the final fault diagnosis model.

[0008] Through the above technical solution, the embodiments of the present application can use the historical vibration data of the steam turbine rotor and the actual state labels corresponding to the historical vibration data to construct a training sample set and a test sample set, thereby training an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set, and then testing the trained initial fault diagnosis model using the test sample set, and further obtaining a final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to diagnose the fault result of the target steam turbine rotor using the final fault diagnosis model, that is, fully extracting the data features of the vibration data using the graph convolutional neural network to deeply mine the relationship between the vibration data and the state of the steam turbine rotor, and improving the accuracy of fault diagnosis.

[0009] Optionally, in an embodiment of the present application, the training of the initial fault diagnosis model constructed by the graph convolutional neural network using the training sample set includes: obtaining any sample data in the training sample set, and slicing the any sample data to obtain a plurality of sub-samples that meet the preset conditions; performing feature transformation on each sub-sample to obtain corresponding time-domain features and frequency-domain features; constructing a graph according to the time-domain features and frequency-domain features of each sub-sample; using the graph convolutional neural network to extract node features from the graph, and classifying and identifying the node features to output a training state label; updating the graph convolutional neural network based on the output state label and the actual state label corresponding to the any sample data until the preset convergence condition is met to obtain the trained initial fault diagnosis model.

[0010] Through the above technical solution, the embodiments of the present application can construct a graph using the data in the training sample set to fully extract the features of the vibration data and effectively mine the non-linear relationship between the state label and the vibration data, and then form a trained initial fault diagnosis model according to the non-linear relationship between the state label and the vibration data.

[0011] Optionally, in an embodiment of the present application, the expression of the initial fault diagnosis model is:

[0012] Y = Softmax(H (2) W 3 + b 3 ),

[0013] where Y is the actual status label, W 3 is the weight matrix in the fully connected layer, b 3 is the bias vector in the fully connected layer, and H (2) is the output value of the second graph convolutional layer.

[0014] Optionally, in an embodiment of the present application, constructing a graph according to the time domain features and frequency domain features of each subsample includes: calculating the similarity between multiple subsamples to obtain a calculation result; assigning weights to each subsample based on the calculation result to obtain an assignment result; using each subsample as a node and constructing the graph in combination with the assignment result.

[0015] Through the above technical solution, the graph constructed in the embodiment of the present application can display the correlation between subsamples to facilitate subsequent mining of feature information.

[0016] Optionally, in an embodiment of the present application, updating the graph convolutional neural network based on the output status label and the actual status label corresponding to any sample data includes: calculating a loss function using the output status label and the actual status label to obtain a calculation result of the loss function; and reversely updating the weights of each layer in the graph convolutional neural network based on the calculation result of the loss function.

[0017] Through the above technical solution, the embodiment of the present application can use the loss function for model training to continuously improve the model accuracy.

[0018] Optionally, in an embodiment of the present application, the expression of the loss function is:

[0019] loss = -[Y log Y′ + (1 - Y) log(1 - Y′)],

[0020] where Y′ is the training status label.

[0021] An embodiment of the second aspect of the present application provides a fault diagnosis method for a steam turbine rotor, which is applied in the model usage stage. The method includes the following steps: obtaining the current vibration signal of the steam turbine rotor to be detected; inputting the current vibration signal into a pre-constructed fault diagnosis model to obtain a fault diagnosis result of the steam turbine rotor to be detected, where the fault diagnosis model is trained by historical vibration data of the steam turbine rotor and the actual status label corresponding to the historical vibration data.

[0022] A fault diagnosis device for a steam turbine rotor according to an embodiment of the third aspect of the present application is applied to the model construction stage. Wherein, the device includes: an acquisition module, configured to acquire historical vibration data of the steam turbine rotor, and construct a training sample set and a test sample set based on the historical vibration data and the actual state labels corresponding to the historical vibration data; a training module, configured to train an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set to obtain a trained initial fault diagnosis model; a test module, configured to test the trained initial fault diagnosis model using the test sample set, generate a final fault diagnosis model of the steam turbine rotor that meets preset test conditions, and use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor.

[0023] Through the above technical solution, the embodiment of the present application can use the historical vibration data of the steam turbine rotor and the actual state labels corresponding to the historical vibration data to construct a training sample set and a test sample set, thereby training an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set, and then testing the trained initial fault diagnosis model with the test sample set, and further obtaining a final fault diagnosis model of the steam turbine rotor that meets preset test conditions, so as to use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor, that is, fully extract the data features of the vibration data by using the graph convolutional neural network to deeply mine the relationship between the vibration data and the state of the steam turbine rotor, and improve the accuracy of fault diagnosis.

[0024] Optionally, in an embodiment of the present application, the training module includes: an acquisition unit, configured to acquire any sample data in the training sample set and slice the any sample data to obtain a plurality of sub-samples that meet preset conditions; a conversion unit, configured to perform feature conversion on each sub-sample to obtain corresponding time-domain features and frequency-domain features; a construction unit, configured to construct a graph according to the time-domain features and frequency-domain features of each sub-sample; an extraction unit, configured to extract node features from the graph using the graph convolutional neural network and perform classification and recognition on the node features to output a training state label; an update unit, configured to update the graph convolutional neural network based on the output state label and the actual state label corresponding to the any sample data until a preset convergence condition is met to obtain the trained initial fault diagnosis model.

[0025] Through the above technical solution, the embodiment of the present application can construct a graph using the data in the training sample set to fully extract the features of the vibration data and effectively mine the non-linear relationship between the state label and the vibration data, and then form a trained initial fault diagnosis model according to the non-linear relationship between the state label and the vibration data.

[0026] Optionally, in an embodiment of the present application, the expression of the initial fault diagnosis model is:

[0027] Y = Softmax(H (2) W 3 + b 3 ),

[0028] where Y is the actual status label, W 3 is the weight matrix in the fully connected layer, b 3 is the bias vector in the fully connected layer, and H (2) is the output value of the second graph convolutional layer.

[0029] Optionally, in an embodiment of the present application, the construction unit includes: a first calculation subunit, configured to calculate the similarity between multiple subsamples to obtain a calculation result; an assignment subunit, configured to assign weights to each of the subsamples based on the calculation result to obtain an assignment result; a construction subunit, configured to use each of the subsamples as a node and construct the graph in combination with the assignment result.

[0030] Through the above technical solution, the graph constructed in the embodiment of the present application can display the correlation between subsamples, so as to mine feature information subsequently.

[0031] Optionally, in an embodiment of the present application, the update unit includes: a second calculation subunit, configured to calculate a loss function using the output status label and the actual status label to obtain a loss function calculation result; an update subunit, configured to reversely update the weights of each layer in the graph convolutional neural network based on the loss function calculation result.

[0032] Through the above technical solution, the embodiment of the present application can use the loss function for model training to continuously improve the model accuracy.

[0033] Optionally, in an embodiment of the present application, the expression of the loss function is:

[0034] loss = -[Y log Y′ + (1 - Y) log(1 - Y′)],

[0035] where Y′ is the training status label.

[0036] An embodiment of the fourth aspect of the present application provides a fault diagnosis device for a steam turbine rotor, which is applied to the model usage stage. The device includes: an acquisition module, configured to acquire the current vibration signal of the steam turbine rotor to be detected; a diagnosis module, configured to input the current vibration signal into a pre-constructed fault diagnosis model to obtain a fault diagnosis result of the steam turbine rotor to be detected, where the fault diagnosis model is trained by historical vibration data of the steam turbine rotor and the actual status label corresponding to the historical vibration data.

[0037] An embodiment of the fifth aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the program to implement the fault diagnosis method of the steam turbine rotor as described in the above embodiment.

[0038] An embodiment of the sixth aspect of the present application provides a computer-readable storage medium storing computer instructions for causing the computer to execute the fault diagnosis method of the steam turbine rotor as described in the above embodiment.

[0039] An embodiment of the seventh aspect of the present application provides a computer program product including a computer program, which when executed, is used to implement the fault diagnosis method of the steam turbine rotor as above.

[0040] Embodiments of the present application can use the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data to construct a training sample set and a test sample set, thereby training an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set, and then testing the trained initial fault diagnosis model with the test sample set, and further obtaining a final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor, and fully extract the data features of the vibration data using the graph convolutional neural network to deeply mine the relationship between the vibration data and the state of the steam turbine rotor, improving the accuracy of fault diagnosis. Thus, it solves the technical problem in the related art that the accuracy of fault diagnosis relying only on similarity measurement is limited, and the model algorithm is difficult to deeply mine the complex relationship between the fault type features and the vibration signal, resulting in a large error in the diagnosis result.

[0041] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present application. Description of the Drawings

[0042] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where:

[0043] Figure 1 It is a flowchart of a fault diagnosis method for a steam turbine rotor according to an embodiment of the present application;

[0044] Figure 2 It is a schematic diagram of the principle of a fault diagnosis method for a steam turbine rotor according to an embodiment of the present application;

[0045] Figure 3 It is a waveform diagram of a steam turbine rotor vibration fault signal and a vibration signal under normal conditions according to an embodiment of the present application;

[0046] Figure 4 is a data processing flow chart according to an embodiment of the present application;

[0047] Figure 5 is a construction diagram flow chart according to an embodiment of the present application;

[0048] Figure 6 is a schematic diagram of the structure of a graph convolutional neural network according to an embodiment of the present application;

[0049] Figure 7 is a schematic diagram of the training loss function of a graph convolutional neural network according to an embodiment of the present application;

[0050] Figure 8 is a schematic diagram of the structure of a fault diagnosis device for a steam turbine rotor provided according to an embodiment of the present application;

[0051] Figure 9 is a flow chart of another fault diagnosis method for a steam turbine rotor provided according to an embodiment of the present application;

[0052] Figure 10 is a schematic diagram of the structure of another fault diagnosis device for a steam turbine rotor provided according to an embodiment of the present application;

[0053] Figure 11 is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application.

[0054] Among them, 10 - fault diagnosis device for steam turbine rotor, 101 - acquisition module, 102 - training module, 103 - testing module; 20 - fault diagnosis device for steam turbine rotor, 201 - acquisition module, 202 - diagnosis module; 1101 - memory, 1102 - processor, 1103 - communication interface. Detailed implementation manners

[0055] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application and should not be construed as limiting the present application.

[0056] The following describes a fault diagnosis method, device, electronic device, and storage medium for a steam turbine rotor according to an embodiment of the present application. In view of the technical problem in the related art mentioned in the above background art that the accuracy of fault diagnosis relying solely on similarity measurement is limited, and the model algorithm is difficult to deeply explore the complex relationship between the fault type characteristics and the vibration signal, resulting in a large error in the diagnosis result, the present application provides a fault diagnosis method for a steam turbine rotor. In this method, the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data can be used to construct a training sample set and a test sample set, so as to train an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set, and then test the trained initial fault diagnosis model with the test sample set, and further obtain the final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor, and use the graph convolutional neural network to fully extract the data characteristics of the vibration data, so as to deeply explore the vibration data and the state of the steam turbine rotor, and improve the accuracy of fault diagnosis. Thus, the technical problem in the related art that the accuracy of fault diagnosis relying solely on similarity measurement is limited, and the model algorithm is difficult to deeply explore the complex relationship between the fault type characteristics and the vibration signal, resulting in a large error in the diagnosis result is solved.

[0057] Specifically, Figure 1 FIG. is a schematic flowchart of a fault diagnosis method for a steam turbine rotor provided by an embodiment of the present application.

[0058] As Figure 1 shown, this fault diagnosis method for a steam turbine rotor is applied to the model construction stage, where the method includes the following steps:

[0059] In step S101, obtain the historical vibration data of the steam turbine rotor, and construct a training sample set and a test sample set based on the historical vibration data and the actual status labels corresponding to the historical vibration data.

[0060] It can be understood that as one of the most important devices in a thermal power plant, a steam turbine has the characteristics of complex structure, harsh working conditions (high temperature, high pressure, high speed), and high requirements for continuous operation, and is prone to failures. Its rotor is even more crucial. The stability of its operation status will immediately affect the power generation system in this area, and then affect the entire power system, causing significant economic losses to the social economy. Therefore, accurately diagnosing the status of the steam turbine rotor is of great significance to the thermal power plant.

[0061] The main faults of a steam turbine rotor include: imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil film oscillation, steam flow excitation, and hybrid faults composed of the above-mentioned faults. Embodiments of the present application can determine the actual state label of the steam turbine rotor and the corresponding data under the state label based on the above-mentioned faults.

[0062] In the actual execution process, embodiments of the present application can obtain the historical vibration data of the steam turbine rotor and the actual state label corresponding to the historical vibration data from the DCS (distributed control system, distributed control system of thermal power plants). For example, four radial vibration data taken from the bearing covers at both ends of the shaft, the labels and data mainly include: steam turbine rotor fault vibration signals such as imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil film oscillation, steam flow excitation, and vibration signals under normal conditions.

[0063] After obtaining the historical vibration data, embodiments of the present application can obtain a sample set by data cleaning using the sliding window method, with the same number of samples for each state, and construct a training sample set and a test sample set.

[0064] For example, embodiments of the present application can set the length of each sample to 4608 data points, including 1152 data points for each radial vibration, and randomly divide them into a training sample set and a test sample set according to a ratio of 7:3.

[0065] In step S102, the initial fault diagnosis model constructed by the graph convolutional neural network is trained using the training sample set to obtain the trained initial fault diagnosis model.

[0066] It should be noted that the graph convolutional neural network has achieved outstanding performance in graph-structured data. In order to deeply explore the relationship between the state of the steam turbine rotor and the vibration signal, embodiments of the present application can construct an initial fault diagnosis model with the graph convolutional neural network as the main body, and use the training sample set to train the initial fault diagnosis model to obtain a fault diagnosis model with high accuracy, so as to realize the fault diagnosis of the steam turbine rotor.

[0067] Optionally, in an embodiment of the present application, training an initial fault diagnosis model constructed by a graph convolutional neural network using a training sample set includes: obtaining any sample data in the training sample set, and slicing the any sample data to obtain a plurality of sub-samples that meet preset conditions; performing feature transformation on each sub-sample to obtain corresponding time-domain features and frequency-domain features; constructing a graph according to the time-domain features and frequency-domain features of each sub-sample; using the graph convolutional neural network to extract node features from the graph, and classifying and identifying the node features to output a training status label; updating the graph convolutional neural network based on the output status label and the actual status label corresponding to the any sample data until a preset convergence condition is met, so as to obtain a trained initial fault diagnosis model.

[0068] As a possible implementation manner, before model training in the embodiment of the present application, data used for training needs to be processed first, that is, the data in the training sample set is processed. In the embodiment of the present application, the vibration signal can be converted into nodes in a graph through data slicing and feature transformation, so that the steam turbine rotor fault detection is transformed into a node classification task in machine learning.

[0069] For example, in the embodiment of the present application, the sample data can be sliced first to divide a large sample into multiple sub-samples. Slicing is performed using a non-overlapping moving window to convert one-dimensional raw data into a two-dimensional matrix. At the same time. Then feature transformation is performed. For each column of data in the two-dimensional matrix, 18 features in the time domain and frequency domain are calculated and used as the input of the subsequent model.

[0070] Furthermore, in the embodiment of the present application, the cleaned data can be used to construct a graph to show the correlation between sub-samples, and the graph convolutional neural network is used to process the graph data to train the initial fault diagnosis model.

[0071] Among them, in the embodiment of the present application, the graph convolutional neural network can be used to extract node features: using the graph convolutional neural network to process graph data. The model of the graph convolutional neural network is based on an information propagation mechanism, in which each node exchanges information with other nodes through continuous iterative updates to reach a stable state. When the information flow of the entire graph is smooth, each node has information about itself and its adjacent nodes, which helps the subsequent fault detection task. After feature extraction by the graph convolutional neural network, each graph node contains not only its own information but also the feature information of its neighbors.

[0072] Optionally, in an embodiment of the present application, constructing a graph according to the time-domain features and frequency-domain features of each sub-sample includes: calculating the similarity between multiple sub-samples to obtain a calculation result; assigning weights to each sub-sample based on the calculation result to obtain an assignment result; using each sub-sample as a node and constructing a graph in combination with the assignment result.

[0073] In some embodiments, a graph can be constructed to show the correlation between subsamples. When constructing the graph, the similarity between subsamples is mainly considered. The higher the similarity, the greater the connection weight between them. Constructing the graph includes three steps: (1) calculating the similarity between subsamples; (2) assigning weights; (3) outputting the graph.

[0074] Optionally, in an embodiment of the present application, the expression of the initial fault diagnosis model is:

[0075] Y = Softmax(H (2) W 3 + b 3 ),

[0076] where Y is the actual status label, W 3 is the weight matrix in the fully connected layer, b 3 is the bias vector in the fully connected layer, and H (2) is the output value of the second graph convolutional layer.

[0077] It should be noted that the graph convolutional neural network is a framework that directly uses deep learning to learn from graph-structured data. In the embodiments of the present application, the graph convolutional neural network is used to process graph data. The model of the graph convolutional neural network is based on an information propagation mechanism, in which each node exchanges information with other nodes through continuous iterative updates to reach a stable state. When the information flow of the entire graph is smooth, each node has information about itself and its adjacent nodes, which helps the subsequent fault detection task. After feature extraction through the graph convolutional neural network, each graph node contains not only its own information but also the feature information of its neighbors.

[0078] The graph-structured data (X, A) is input into the graph convolutional neural network, where X is the feature data sample and A is the adjacency matrix. The graph convolutional neural network outputs H (1) on the first layer:

[0079] H (1) = ReLU(AXW 1 + b 1 ),

[0080] where W 1 and b 1 are respectively the weight matrix and the bias vector in the first graph convolutional layer.

[0081] Similarly, the output data of the second graph convolutional layer is:

[0082] H (2) = ReLU(AH (1) W 2 + b 2 ),

[0083] Among them, W 2 and b 2 are respectively the weight matrix and the bias vector in the second graph convolutional layer.

[0084] Behind the graph convolutional layer is a fully connected layer. The Softmax function output is:

[0085] Y = Softmax(H (2) W 3 + b 3 ),

[0086] Among them, W 3 and b 3 are respectively the weight matrix and the bias vector in the fully connected layer. Y is the label corresponding to the state of the steam turbine rotor.

[0087] Among them, ReLU (Rectified Linear Unit) is an activation function, which can enhance the non-linear expression ability of the model. Softmax is a normalization function, which is used to map the input vector into the probabilities of each category, so as to form a probability distribution, ensure that each element of the output is between (0, 1), and the sum of all elements is 1.

[0088] Optionally, in an embodiment of the present application, updating the graph convolutional neural network based on the output state label and the actual state label corresponding to any sample data includes: calculating a loss function using the output state label and the actual state label to obtain a loss function calculation result; updating the weights of each layer in the graph convolutional neural network in reverse based on the loss function calculation result. Among them, the expression of the loss function is:

[0089] loss = -[YlogY′ + (1 - Y)log(1 - Y′)],

[0090] where Y′ is the training state label.

[0091] Embodiments of the present application can use a training set to train the model, and update the learnable parameters of the model: W and b through the error backpropagation algorithm, mainly including two steps: (1) forward excitation propagation; (2) reverse weight update. For forward excitation propagation, embodiments of the present application can input variables to be processed by multiple graph convolutional layers, and then transfer them to the fully connected layer, and the labels of the samples are output by the fully connected layer. The diagnostic result and the actual result are used to calculate the loss function (error). For reverse weight update, embodiments of the present application can use the chain rule to transfer the error from the output layer to the intermediate layer. Then, update the weights of each layer through the gradient descent method. Repeat these steps until the loss function converges or reaches a preset number of iterations to complete the training process.

[0092] Among them, the loss function can be a cross-entropy loss function.

[0093] In step S103, the trained initial fault diagnosis model is tested using the test sample set to generate the final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to diagnose the fault result of the target steam turbine rotor using the final fault diagnosis model.

[0094] After obtaining the trained initial fault diagnosis model, the embodiments of the present application can perform model testing using the test sample set. Among them, the test sample set also needs to go through the same data processing method as the training sample set.

[0095] The embodiments of the present application can obtain the test result of the model, that is, input the vibration signal in the test sample set into the trained initial fault diagnosis model to obtain the diagnosis result, that is, the test status label.

[0096] By comparing the test status label with the corresponding actual status label in the test sample set to determine whether the trained initial fault diagnosis model passes the test, and after passing the test, obtain the final fault diagnosis model for fault diagnosis of the target steam turbine rotor.

[0097] Combined with Figures 2 to 7 shown, the working principle of the fault diagnosis method for the steam turbine rotor in the embodiments of the present application is elaborated in detail with an example.

[0098] As Figure 2 shown, the embodiments of the present application may include the following steps:

[0099] Step S1: Obtaining the historical data set of the steam turbine rotor.

[0100] As Figure 3 shown, it is the waveform diagram of the vibration signal and 9 common fault signals of the steam turbine rotor under normal conditions. The main faults of the steam turbine rotor include: imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil whip, steam-excited vibration, etc. The differences between them are manifested in signal amplitude and spectrum.

[0101] Step S2: Data processing.

[0102] As Figure 4 shown, it is the process of processing the vibration signal of the steam turbine rotor, which is converted from one-dimensional raw data to a two-dimensional matrix. The sample data usually used for fault detection of the steam turbine rotor is N*1 time series data, and the sample data is sliced using a non-overlapping moving window. The sample data is cut into segments containing 384 data points, and then the original sample data is converted into a 384*(N / 384) two-dimensional matrix. The converted sample data contains a total of N / 384 sub-samples, and each sub-sample consists of 384 data points.

[0103] For example, the sample data for steam turbine rotor fault detection is 4608 * 1 time series data, and the sample data is sliced using a non-overlapping moving window. The sample data is cut into segments containing 384 data points, and then the original sample data is converted into a 384 * 12 matrix. The converted sample data contains a total of 12 sub-samples, and each sub-sample consists of 384 data points.

[0104] Next, feature transformation is performed. Through feature transformation, redundant information in the sub-samples can be reduced, effectively reducing the computational complexity of the subsequent model. For each sub-sample, 18 features in the time domain and frequency domain are calculated and used as the input of the subsequent model.

[0105] (1) The calculation formulas for 10 time domain features are as follows:

[0106] Mean value (mean):

[0107] Standard deviation:

[0108] Root mean square value:

[0109] Peak value (peak): y 4 = max(|x(i)|)

[0110] Skewness:

[0111] Kurtosis:

[0112] Crest factor:

[0113] Clearance Factor:

[0114] Shape factor:

[0115] Impulse factor:

[0116] Where x(i) is the i-th data in the sub-sample, and N is the total number of data in the sub-sample.

[0117] Eight frequency domain features are obtained by three-layer wavelet packet decomposition of eight frequency bands of wavelet energy. The wavelet function is selected as Db4, and the calculation formula is:

[0118]

[0119] Among them, n = 0, 1, …, 2j - 1, where j is the wavelet packet decomposition level, and w n (i) is the i-th wavelet coefficient of the n-th decomposition level.

[0120] Step S3: Construct a graph.

[0121] As Figure 5 shown, for the process of constructing a graph, in the embodiments of the present application, a graph can be constructed to show the correlation between subsamples. When constructing a graph, the similarity between subsamples is mainly considered. The higher the similarity, the greater the connection weight between them. Constructing a graph includes three steps: (1) calculating the similarity between subsamples; (2) assigning weights; (3) outputting the graph.

[0122] Among them, (1) in the embodiments of the present application, the similarity between subsamples can be calculated. Let X = {X 1 , X 2 , X 3 , …, X m} represent the processed feature data samples, where m represents the number of subsamples, X i is x i1 , x i2 , x i3 , …, x in representing the i-th subsample in X, n represents the dimension of X, and X in represents the value of the subsample X i in the n-th dimension. The normalized Euclidean distance is used as a measure of the similarity between subsamples. Let Dist(X i , X j ) represent the similarity between subsamples X i and X j , and the calculation formula is:

[0123]

[0124] When the value of Dist(X i , X j ) is relatively large, it means that the similarity between the two objects is relatively high.

[0125] (2) Assign weights. First, in the embodiments of the present application, the top k subsamples with the highest similarity to the X i subsample can be selected to form the neighbor set of X i , denoted as N k (X i ). Calculate the weights between these k subsamples and X i , and the calculation formula is:

[0126]

[0127] (3) Output graph. Each subsample (a group of feature vectors x i1 , x i2 , x i3 , …, x in ) can be regarded as a node of the graph model. In the embodiments of the present application, after calculating the similarity and assigning weights in the first two steps, the subsamples can be connected to each other. If the weight between subsamples is greater than 0, there is a connecting edge between them; otherwise, there is no connection. These subsamples are connected to form a graph through the adjacency matrix A, and the adjacency matrix A is:

[0128]

[0129] The values of the diagonal elements in the adjacency matrix A are all 1, which indicates that the subsample is connected to itself. In this way, the characteristic information of the subsample itself can be effectively prevented from being lost during the training process of the subsequent model.

[0130] Step S4: Extract node features.

[0131] As Figure 6 shown, the graph convolutional neural network is a framework that uses deep learning to directly learn from graph-structured data. In the embodiments of the present application, the graph convolutional neural network is used to process graph data. The model of the graph convolutional neural network is based on an information propagation mechanism, in which each node exchanges information with other nodes through continuous iterative updates to reach a stable state. When the information flow of the entire graph is smooth, each node has information about itself and its adjacent nodes, which helps the subsequent fault detection task. After feature extraction through the graph convolutional neural network, each graph node not only contains its own information but also contains the feature information of its neighbors.

[0132] The graph-structured data (X, A) is input into the graph convolutional neural network, where X is the feature data sample and A is the adjacency matrix. The graph convolutional neural network outputs H (1) :

[0133] H (1) = ReLU(AXW 1 + b 1 ),

[0134] where W 1 and b 1 are the weight matrix and bias vector in the first graph convolutional layer respectively.

[0135] Similarly, the output data of the second graph convolutional layer is:

[0136] H (2) = ReLU(AH (1) W2 +b 2 ),

[0137] Among them, W 2 and b 2 are respectively the weight matrix and the bias vector in the second graph convolutional layer.

[0138] Behind the graph convolutional layer is a fully connected layer. The Softmax function output is:

[0139] Y = Softmax(H (2) W 3 +b 3 ),

[0140] Among them, W 3 and b 3 are respectively the weight matrix and the bias vector in the fully connected layer. Y is the label corresponding to the steam turbine rotor state.

[0141] Among them, ReLU (Rectified Linear Unit) is the activation function, which can enhance the non-linear expression ability of the model. Softmax is the normalization function, which is used to map the input vector to the probabilities of each category, so as to form a probability distribution, ensuring that each element of the output is between (0, 1) and the sum of all elements is 1.

[0142] The embodiments of the present application can use the training set to train the model, and update the learnable parameters of the model: W and b through the error backpropagation algorithm, which mainly includes two steps: (1) Forward excitation propagation; (2) Backward weight update. For forward excitation propagation, the input variables in the embodiments of the present application can be processed by multiple graph convolutional layers and then transferred to the fully connected layer, and the labels of the samples are output by the fully connected layer. The diagnostic results and the actual results are used to calculate the loss function (error). For backward weight update, the embodiments of the present application can use the chain rule to transfer the error from the output layer to the intermediate layer. Then, the weights of each layer are updated by the gradient descent method. Repeat these steps until the loss function converges or reaches the preset number of iterations to complete the training process.

[0143] Among them, the loss function can be the cross-entropy loss function, and the expression of the loss function is:

[0144] loss = -[YlogY′ + (1 - Y)log(1 - Y′)],

[0145] Among them, Y′ is the training status label.

[0146] The change trend of the loss function with the increase of the number of iterations can be as Figure 7As shown. In the early stage of the training process, the loss function of the training sample set decreases rapidly as the number of iterations increases. When the number of iterations is greater than 400, the loss function tends to be constant and no longer continues to decrease, indicating that the model has converged. To ensure the convergence of the model, the model can be used for diagnosing unknown samples after 800 iterations.

[0147] Step S5: Fault diagnosis.

[0148] The embodiment of the present application can use a fully connected layer and Softmax as a classifier to classify and identify the extracted node features, and output the fault diagnosis results, including the normal working state, and fault types such as imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil film oscillation, steam flow excitation, etc.

[0149] For example, the output of the model of the embodiment of the present application is the label corresponding to the state of the steam turbine rotor (10000 00000 represents normal, 01000 00000 represents imbalance, 00100 00000 represents seal friction, 00010 00000 represents rotor axial friction, 00001 00000 represents misalignment, 00000 10000 represents journal and bearing eccentricity, 00000 01000 represents casing deformation, 00000 00100 represents bearing damage, 00000 00010 represents oil film oscillation, 00000 00001 represents steam flow excitation).

[0150] The embodiment of the present application can use the test samples to construct a graph and an adjacency matrix through data slicing and feature transformation as the input of the graph convolutional neural network, extract the fault features of the steam turbine rotor through the trained model, and then use a fully connected layer and Softmax as a classifier to classify and identify the extracted node features, and output the fault diagnosis results, including the normal working state, and fault types such as imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil film oscillation, steam flow excitation, etc.

[0151] In summary, the embodiments of the present application can preprocess the original vibration signals of the steam turbine rotor, and convert the time series signals of vibration into non-Euclidean structure diagram data through methods such as data slicing, feature transformation, and similarity measurement; the final fault diagnosis model of the embodiments of the present application can use a graph convolutional neural network to extract the features of the vibration signals, and input the constructed graph and the adjacency matrix into the graph convolutional neural network for processing. After the processing is completed, each graph node not only contains its own information, but also contains the feature information of its neighbors; the embodiments of the present application can accurately complete the fault diagnosis of various types of steam turbine rotors, including fault states such as imbalance, seal friction, rotor axial friction, misalignment, journal and bearing eccentricity, casing deformation, bearing damage, oil whip, and steam flow excitation, and have a higher fault recognition rate compared with methods such as CNN, ANN, SVM, and K-nearest neighbor.

[0152] According to the fault diagnosis method of the steam turbine rotor proposed by the embodiments of the present application, the historical vibration data of the steam turbine rotor and the actual state labels corresponding to the historical vibration data can be used to construct a training sample set and a test sample set, so as to train an initial fault diagnosis model constructed by a graph convolutional neural network using the training sample set, and then test the trained initial fault diagnosis model with the test sample set, and further obtain the final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, so as to use the final fault diagnosis model to diagnose the fault results of the target steam turbine rotor, and use the graph convolutional neural network to fully extract the data features of the vibration data, so as to deeply explore the vibration data and the state of the steam turbine rotor, and improve the accuracy of fault diagnosis. Thus, the technical problem in the related art that the accuracy of fault diagnosis relying only on similarity measurement is limited, and the model algorithm is difficult to deeply explore the complex relationship between the fault type features and the vibration signals, resulting in a large error in the diagnosis results is solved.

[0153] Next, a fault diagnosis device for a steam turbine rotor according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0154] Figure 8 It is a block diagram of a fault diagnosis device for a steam turbine rotor according to an embodiment of the present application.

[0155] As Figure 8 shown, the fault diagnosis device 10 for the steam turbine rotor is applied to the model construction stage, wherein the device 10 includes: an acquisition module 101, a training module 102, and a testing module 103.

[0156] Specifically, the acquisition module 101 is configured to acquire the historical vibration data of the steam turbine rotor, and construct a training sample set and a testing sample set based on the historical vibration data and the actual state labels corresponding to the historical vibration data.

[0157] The training module 102 is used to train an initial fault diagnosis model constructed by a graph convolutional neural network using a training sample set to obtain a trained initial fault diagnosis model.

[0158] The testing module 103 is used to test the trained initial fault diagnosis model using a test sample set, generate a final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, and use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor.

[0159] Optionally, in an embodiment of the present application, the training module 102 includes: an acquisition unit, a conversion unit, a construction unit, an extraction unit, and an update unit.

[0160] Among them, the acquisition unit is used to acquire any sample data in the training sample set and slice any sample data to obtain a plurality of sub-samples that meet the preset conditions.

[0161] The conversion unit is used to perform feature conversion on each sub-sample to obtain corresponding time-domain features and frequency-domain features.

[0162] The construction unit is used to construct a graph according to the time-domain features and frequency-domain features of each sub-sample.

[0163] The extraction unit is used to extract node features from the graph using a graph convolutional neural network and perform classification and recognition on the node features to output a training status label.

[0164] The update unit is used to update the graph convolutional neural network based on the output status label and the actual status label corresponding to any sample data until the preset convergence condition is met to obtain a trained initial fault diagnosis model.

[0165] Optionally, in an embodiment of the present application, the expression of the initial fault diagnosis model is:

[0166] Y = Softmax(H (2) W 3 + b 3 ),

[0167] Among them, Y is the actual status label, W 3 is the weight matrix in the fully connected layer, b 3 is the bias vector in the fully connected layer, and H (2) is the output value of the second graph convolutional layer.

[0168] Optionally, in an embodiment of the present application, the construction unit includes: a first calculation sub-unit, an assignment sub-unit, and a construction sub-unit.

[0169] Among them, the first calculation sub-unit is used to calculate the similarity between a plurality of sub-samples to obtain a calculation result.

[0170] An allocation subunit, configured to allocate weights to each subsample based on a calculation result to obtain an allocation result.

[0171] A construction subunit, configured to use each subsample as a node and construct a graph in combination with the allocation result.

[0172] Optionally, in an embodiment of the present application, the update unit includes: a second calculation subunit and an update subunit.

[0173] Wherein, the second calculation subunit is configured to calculate a loss function by using an output state label and an actual state label to obtain a calculation result of the loss function.

[0174] The update subunit is configured to reversely update the weights of each layer in the graph convolutional neural network based on the calculation result of the loss function.

[0175] Optionally, in an embodiment of the present application, the expression of the loss function is:

[0176] loss = -[YlogY′+(1 - Y)log(1 - Y′)],

[0177] Wherein, Y′ is a training state label.

[0178] It should be noted that the foregoing explanation of the embodiment of the fault diagnosis method for a steam turbine rotor also applies to the fault diagnosis device for a steam turbine rotor in this embodiment, and details are not described herein again.

[0179] According to the fault diagnosis device for a steam turbine rotor provided by an embodiment of the present application, historical vibration data of the steam turbine rotor and actual state labels corresponding to the historical vibration data can be used to construct a training sample set and a test sample set, so as to train an initial fault diagnosis model constructed by a graph convolutional neural network by using the training sample set, and then test the trained initial fault diagnosis model by using the test sample set, and further obtain a final fault diagnosis model for the steam turbine rotor that meets preset test conditions, so as to use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor, and fully extract the data features of the vibration data by using the graph convolutional neural network to deeply excavate the relationship between the vibration data and the state of the steam turbine rotor, and improve the accuracy of fault diagnosis. Thus, the technical problem in the related art that the accuracy of fault diagnosis relying only on similarity measurement is limited, and the model algorithm is difficult to deeply excavate the complex relationship between the fault type features and the vibration signal, resulting in a large error in the diagnosis result, is solved.

[0180] The above is the elaboration of the embodiment of the present application in the model construction stage. The following describes the model usage stage of the embodiment of the present application.

[0181] Specifically, Figure 9Schematic flowchart of a fault diagnosis method for a steam turbine rotor provided by an embodiment of the present application.

[0182] As Figure 9 shown, this fault diagnosis method for a steam turbine rotor is applied in the model usage stage. Among them, the method includes the following steps:

[0183] In step S901, obtain the current vibration signal of the steam turbine rotor to be detected.

[0184] In step S902, input the current vibration signal into a pre-constructed fault diagnosis model to obtain the fault diagnosis result of the steam turbine rotor to be detected. Among them, the fault diagnosis model is trained by the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data.

[0185] According to the fault diagnosis method for a steam turbine rotor proposed by an embodiment of the present application, a training sample set and a test sample set can be constructed by using the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data. Then, the initial fault diagnosis model constructed by a graph convolutional neural network is trained by using the training sample set, and the trained initial fault diagnosis model is tested by using the test sample set. Furthermore, the final fault diagnosis model of the steam turbine rotor that meets the preset test conditions is obtained, so as to use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor. The graph convolutional neural network is used to fully extract the data features of the vibration data, so as to deeply explore the relationship between the vibration data and the state of the steam turbine rotor, and improve the accuracy of fault diagnosis. Thus, the technical problem in the related art that the accuracy of fault diagnosis relying only on similarity measurement is limited, and the model algorithm is difficult to deeply explore the complex relationship between the fault type features and the vibration signal, resulting in a large error in the diagnosis result is solved.

[0186] Secondly, a fault diagnosis device for a steam turbine rotor proposed by an embodiment of the present application is described with reference to the accompanying drawings.

[0187] Figure 10 is a block diagram of a fault diagnosis device for a steam turbine rotor according to an embodiment of the present application.

[0188] As Figure 10 shown, this fault diagnosis device 20 for a steam turbine rotor is applied in the model usage stage. Among them, the device 20 includes: an acquisition module 201 and a diagnosis module 202.

[0189] Specifically, the acquisition module 201 is used to obtain the current vibration signal of the steam turbine rotor to be detected.

[0190] The diagnostic module 202 is configured to input the current vibration signal into a pre-constructed fault diagnosis model to obtain the fault diagnosis result of the steam turbine rotor to be detected, wherein the fault diagnosis model is trained by the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data.

[0191] It should be noted that the foregoing explanation of the embodiments of the fault diagnosis method for the steam turbine rotor also applies to the fault diagnosis device for the steam turbine rotor in this embodiment, and will not be elaborated here.

[0192] According to the fault diagnosis device for the steam turbine rotor provided by the embodiments of the present application, a training sample set and a test sample set can be constructed by using the historical vibration data of the steam turbine rotor and the actual status labels corresponding to the historical vibration data. Thus, the initial fault diagnosis model constructed by the graph convolutional neural network can be trained by using the training sample set, and then the trained initial fault diagnosis model can be tested by using the test sample set. Furthermore, the final fault diagnosis model of the steam turbine rotor that meets the preset test conditions can be obtained to diagnose the fault result of the target steam turbine rotor by using the final fault diagnosis model. The data features of the vibration data are fully extracted by using the graph convolutional neural network to deeply mine the relationship between the vibration data and the state of the steam turbine rotor, thereby improving the accuracy of fault diagnosis. Therefore, the technical problem in the related art that the accuracy of fault diagnosis relying only on similarity measurement is limited, and the model algorithm is difficult to deeply mine the complex relationship between the fault type features and the vibration signal, resulting in a large error in the diagnosis result, is solved.

[0193] Figure 11 The following is a schematic structural diagram of the electronic device provided by the embodiments of the present application. The electronic device may include:

[0194] A memory 1101, a processor 1102, and a computer program stored on the memory 1101 and executable on the processor 1102.

[0195] When the processor 1102 executes the program, it implements the fault diagnosis method for the steam turbine rotor provided in the foregoing embodiments.

[0196] Further, the electronic device further includes:

[0197] A communication interface 1103 for communication between the memory 1101 and the processor 1102.

[0198] The memory 1101 is used to store the computer program executable on the processor 1102.

[0199] The memory 1101 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory.

[0200] If the memory 1101, the processor 1102, and the communication interface 1103 are implemented independently, the communication interface 1103, the memory 1101, and the processor 1102 can be interconnected via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 11 only a thick line is used to represent it in Figure 11 , but it does not mean that there is only one bus or one type of bus.

[0201] Optionally, in a specific implementation, if the memory 1101, the processor 1102, and the communication interface 1103 are integrated on a single chip, the memory 1101, the processor 1102, and the communication interface 1103 can communicate with each other through an internal interface.

[0202] The processor 1102 may be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0203] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the fault diagnosis method of the steam turbine rotor as described above is implemented.

[0204] The embodiments of the present application also provide a computer program product, including a computer program. When the computer program is executed by a processor, the fault diagnosis method of the steam turbine rotor provided by the embodiments of the present invention is implemented.

[0205] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of this 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 may be combined in any one or N embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0206] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of this application, the meaning of "N" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined.

[0207] Any process or method description shown in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or N executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of this application includes additional implementations, where the functions may be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of this application belong.

[0208] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definable sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or N wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then stored in a computer memory.

[0209] It should be understood that various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0210] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the method of implementing the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0211] In addition, each functional unit in various embodiments of the present application may be integrated into a processing module, may exist physically alone for each unit, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0212] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for diagnosing a fault of a steam turbine rotor, characterized in that: Applied to the model building stage, wherein the method comprises the following steps: Acquire historical vibration data of a steam turbine rotor, and construct a training sample set and a test sample set based on the historical vibration data and actual state labels corresponding to the historical vibration data; Using the training sample set to train an initial fault diagnosis model constructed by a graph convolutional neural network to obtain a trained initial fault diagnosis model; The trained initial fault diagnosis model is tested using the test sample set to generate a final fault diagnosis model of the steam turbine rotor that meets preset test conditions, so as to diagnose the fault result of the target steam turbine rotor using the final fault diagnosis model.

2. The method according to claim 1, characterized in that: The using the training sample set to train the initial fault diagnosis model constructed by the graph convolutional neural network includes: Obtaining any sample data in the training sample set, and slicing the any sample data to obtain a plurality of sub-samples that meet preset conditions; Perform feature transformation on each sub-sample to obtain corresponding time domain features and frequency domain features; constructing a graph according to the time domain characteristics and the frequency domain characteristics of each sub-sample; Extracting node features from the graph using the graph convolutional neural network, and classifying and identifying the node features to output training state labels; The graph convolutional neural network is updated based on the output state label and the actual state label corresponding to any sample data until a preset convergence condition is met to obtain the trained initial fault diagnosis model.

3. The method according to claim 2, characterized in that The expression of the initial fault diagnosis model is: <h2 style=";text-align:left;direction:ltr">Y = Softmax(H<h2 style=";text-align:left;direction:ltr"> (2) <h2 style=";text-align:left;direction:ltr"> W3+b3), Among them, Y is the actual state label, W3 is the weight matrix in the fully connected layer, b3 is the bias vector in the fully connected layer, and H (2) is the output value of the second graph convolutional layer.

4. The method according to claim 2, characterized in that: The constructing a graph according to the time domain characteristics and frequency domain characteristics of each sub-sample includes: Calculate the similarity between multiple sub-samples and obtain the calculation result; Based on the calculation result, a weight is assigned to each sub-sample to obtain an assignment result; Each of the sub-samples is taken as a node, and the graph is constructed in combination with the allocation result.

5. The method according to claim 2, characterized in that: The updating of the graph convolutional neural network based on the output state label and the actual state label corresponding to any sample data includes: Calculating a loss function using the output state label and the actual state label to obtain a loss function calculation result; The weights of each layer in the graph convolutional neural network are reversely updated based on the calculation result of the loss function.

6. The method according to claim 5, characterized in that The expression of the loss function is: loss=-[Y log Y′+(1-Y)log(1-Y′)], Among them, Y′ is the training state label.

7. A method for diagnosing a fault of a steam turbine rotor, characterized in that: The steam turbine rotor fault diagnosis method according to any one of claims 1 to 6 is applied to the model use stage, wherein the method comprises the following steps: Acquire the current vibration signal of the steam turbine rotor to be detected; The current vibration signal is input into a pre-built fault diagnosis model to obtain a fault diagnosis result of the turbine rotor to be detected, wherein the fault diagnosis model is trained by historical vibration data of the turbine rotor and actual state labels corresponding to the historical vibration data.

8. A fault diagnosis device for a steam turbine rotor, characterized in that: Applied to the model building stage, wherein the device comprises: An acquisition module, used for acquiring historical vibration data of a steam turbine rotor, and constructing a training sample set and a test sample set based on the historical vibration data and actual state labels corresponding to the historical vibration data; A training module, used to train the initial fault diagnosis model constructed by the graph convolutional neural network using the training sample set to obtain a trained initial fault diagnosis model; The testing module is used to test the trained initial fault diagnosis model using the test sample set, generate a final fault diagnosis model of the steam turbine rotor that meets the preset test conditions, and use the final fault diagnosis model to diagnose the fault result of the target steam turbine rotor.

9. A fault diagnosis device for a steam turbine rotor, characterized in that: Applied to the model use stage, wherein the device comprises: An acquisition module, used for acquiring a current vibration signal of a steam turbine rotor to be detected; A diagnostic module is used to input the current vibration signal into a pre-built fault diagnosis model to obtain a fault diagnosis result of the turbine rotor to be detected, wherein the fault diagnosis model is trained by historical vibration data of the turbine rotor and actual state labels corresponding to the historical vibration data.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the fault diagnosis method for a steam turbine rotor as described in any one of claims 1-6 or 7.

Images