Fault diagnosis method based on steam turbine generator stator end vibration

Abstract

The invention discloses a steam turbine generator stator end vibration-based fault diagnosis method, which comprises the steps of taking images generated by different vibration signal conversion methods as input, extracting features by using VGG-GAM and classifying the features, thereby realizing accurate fault diagnosis, solving the problem that generator fault diagnosis depends too much on experience of operators, and improving the fault diagnosis accuracy. A reliable decision basis is provided for fault maintenance of the generator, and the reliability and the accuracy of information acquisition are improved.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a fault diagnosis method based on the vibration of a stator end of a steam turbine generator. Background Art

[0002] Generators play a vital role in the power system. The stability and reliability of their operation are directly related to the continuity and safety of power supply. With the development of modern industry, higher requirements are placed on the efficient and reliable operation of generators. However, during the long-term operation of generators, they will be affected by mechanical wear, electrical failures, environmental factors and other factors, which may lead to various faults, such as bearing failure, winding short circuit, rotor imbalance, etc. If these faults cannot be detected and handled in time, they will not only cause equipment damage, but may also cause more serious safety accidents.

[0003] However, the existing technology has the following defects and deficiencies: First, fault diagnosis depends on expert experience: Traditional generator fault diagnosis mainly relies on experienced experts to make judgments. Although this method is effective to a certain extent, it has the problems of strong subjectivity and difficulty in standardization. For complex faults, the diagnosis results may vary from person to person, and inexperienced personnel may not be able to accurately identify early faults. Second, the limitations of signal processing technology: Traditional signal processing methods, such as Fourier transform, time-frequency analysis, etc., although they can extract certain fault features, have limited processing capabilities for nonlinear and non-stationary signals. In practical applications, generator fault signals often have complex time-varying and nonlinear characteristics, and traditional methods are difficult to effectively capture these characteristics, resulting in the accuracy of fault diagnosis being affected. Third, the limitations of data-driven methods: In recent years, methods based on machine learning and deep learning have been widely used in the field of fault diagnosis. These methods rely on a large amount of labeled data for training, but in practical applications, fault data is often scarce, especially data on some rare faults is relatively rare, resulting in insufficient model training. In addition, data-driven methods rely heavily on data quality, and noise and abnormal data may lead to a decrease in the robustness of the model. Fourth, challenges of online monitoring and real-time diagnosis: Real-time monitoring of generator operating status and online diagnosis of faults are important requirements of modern power systems for fault diagnosis technology. However, existing technologies still face challenges in real-time and processing capabilities. Real-time monitoring systems need to process a large amount of real-time data, and existing fault diagnosis methods often find it difficult to meet the requirements of real-time processing while ensuring accuracy. Fifth, the integration and intelligence of fault diagnosis systems are insufficient: Many existing fault diagnosis systems have the problem of isolated operation and are difficult to effectively integrate with other systems. In addition, the existing systems have limited intelligence, lack self-learning and adaptive capabilities, and are difficult to cope with complex and changing operating environments.

[0004] In general, the existing generator fault diagnosis technology still has many shortcomings in terms of accuracy, real-time, robustness and intelligence. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a fault diagnosis method based on the vibration of the stator end of a steam turbine generator to solve the problems raised in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a fault diagnosis method based on the vibration of the stator end of a steam turbine generator, comprising:

[0008] Collecting vibration information of the stator end of the steam turbine generator, filtering out noise of the vibration information and normalizing the vibration information to create a vibration signal data set;

[0009] The MTF-GAF feature image conversion method is used on the vibration signal data set to generate three feature images converted from the vibration signals;

[0010] The transformed feature images are divided into a training set and a test set. The feature images of the training set are input into the VGG-GAM model through multi-channels for training to obtain the best model. The feature images of the test set are then input into the best model for fault diagnosis, and the fault type is output to form a fault diagnosis method for the stator end vibration of the steam turbine generator.

[0011] As a preferred solution of the fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to the present invention, the step of filtering out the noise of the vibration information includes:

[0012] The empirical mode decomposition method is used to decompose the original vibration signal into multiple intrinsic mode function components and a residual term;

[0013] By analyzing the frequency spectrum characteristics or statistical indicators of each intrinsic mode function component, the noise content in each intrinsic mode function component is determined;

[0014] Selecting a noise removal or noise suppression processing method according to the noise content;

[0015] The intrinsic mode function component after noise removal or suppression is combined with the residual term to reconstruct the filtered vibration signal.

[0016] As a preferred solution of the fault diagnosis method based on the vibration of the stator end of the steam turbine generator described in the present invention, wherein: a processing method for removing or suppressing noise is selected according to the noise content, including:

[0017] If the noise is to be removed, the intrinsic mode function components that are judged to be noise are directly discarded;

[0018] If the processing is noise suppression, threshold processing is applied to the intrinsic mode function components containing some noise to suppress the noise while retaining the key vibration signal.

[0019] As a preferred solution of the fault diagnosis method based on the vibration of the stator end of the steam turbine generator described in the present invention, the vibration information is normalized, including:

[0020] The vibration information without noise filtered out is normalized, and the range of the vibration information without noise filtered out is compressed to [0, 1].

[0021] As a preferred solution of the fault diagnosis method based on the stator end vibration of the steam turbine generator described in the present invention, wherein: the MTF-GAF feature image conversion method is used for the vibration signal data set to generate three feature images converted from the vibration signal, including:

[0022] By discretizing the state of the vibration signal and capturing the transition probability between the states of the vibration signal, the one-dimensional vibration signal is converted into a two-dimensional matrix form;

[0023] A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the cosine value of the angle sum;

[0024] A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the angle difference.

[0025] As a preferred solution of the fault diagnosis method based on the stator end vibration of the steam turbine generator described in the present invention, wherein: by discretizing the state of the vibration signal and capturing the transition probability between the states of the vibration signal, the one-dimensional vibration signal information is converted into a two-dimensional matrix form, including:

[0026] Normalize the vibration signal, and discretize the normalized vibration signal into Q states;

[0027] According to the discretized vibration signal, the transition probability between Q states is calculated to generate a state transfer matrix. By traversing the entire vibration signal, the number of transitions from state i to state j is counted, and the transition probability of each state is calculated to generate a state transfer matrix.

[0028] The state transfer matrix is ​​used to construct a Markov transition field matrix.

[0029] As a preferred solution of the fault diagnosis method based on the stator end vibration of the steam turbine generator described in the present invention, wherein: a two-dimensional image is generated by mapping a one-dimensional vibration signal to an angle space and calculating the cosine value of the angle sum, including:

[0030] Perform angle mapping on the normalized vibration signal, converting each point in the vibration signal into a corresponding angle;

[0031] The angles obtained by mapping are used to calculate the cosine value of the angle sum of every two time points and construct the Gram sum matrix.

[0032] As a preferred solution of the fault diagnosis method based on the vibration of the stator end of the steam turbine generator described in the present invention, wherein: a one-dimensional vibration signal is mapped to an angle space, and an angle difference is calculated to generate a two-dimensional image, including:

[0033] Perform angle mapping on the normalized vibration signal, converting each point in the vibration signal into a corresponding angle;

[0034] The angle obtained by mapping is used to calculate the sine value of the angle difference between every two time points to construct the Gram difference matrix.

[0035] As a preferred solution of the fault diagnosis method based on the stator end vibration of the steam turbine generator described in the present invention, the VGG-GAM model includes:

[0036] Based on the VGG network structure, the VGG network architecture consists of multiple VGG convolution blocks, each of which includes multiple convolution layers and a maximum pooling layer;

[0037] The input layer receives three feature images generated by the vibration signal transformation, wherein the size of the feature images is M×M×1, and the three feature images are normalized and merged into an input tensor with three channels, wherein the size is M×M×3;

[0038] The GAM module is introduced after the convolution layer to extract the global information of the convolution feature map through average pooling and maximum pooling. The pooling results are concatenated and then the attention map is generated through convolution operation and added to the original feature map.

[0039] The feature extraction part of the model consists of multiple VGG convolution blocks and the introduced GAM module. After the feature extraction is completed, the final feature map is flattened into a one-dimensional vector, passed through two fully connected layers, and the fault category is output using the Softmax classifier.

[0040] As a preferred solution of the fault diagnosis method based on the stator end vibration of the steam turbine generator described in the present invention, it also includes:

[0041] The Kaiming method is used to initialize the convolutional layer weights.

[0042] Compared with the prior art, the invention has the following beneficial effects:

[0043] 1. By capturing different dimensional features of vibration signals through MTF (Markov Transformation Field), GASF (Gram Angle Field) and GADF (Gram Angle Difference Field), a more comprehensive and detailed signal feature expression can be provided, thereby improving the accuracy and robustness of fault diagnosis;

[0044] 2. A fault vibration model based on VGG-GAM was constructed, and the traditional VGG model was improved so that the model paid more attention to important image features and ignored noise and irrelevant areas, thereby improving the accuracy of fault classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0046] Figure 1 The overall flow chart of a fault diagnosis method based on the vibration of the stator end of a steam turbine generator according to an embodiment of the present invention;

[0047] Figure 2 A system architecture diagram of a fault diagnosis method based on stator end vibration of a steam turbine generator according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0051] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0052] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0053] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] Example 1

[0055] Reference Figure 1 , which is the first embodiment of the present invention, provides a fault diagnosis method based on the vibration of the stator end of a steam turbine generator, comprising:

[0056] S1, collecting vibration information of the stator end of the steam turbine generator, filtering out the noise of the vibration information and normalizing the vibration information to create a vibration signal data set;

[0057] Furthermore, the empirical mode decomposition method is used to decompose the original vibration signal x(t) into multiple intrinsic mode function components IMF 1 ,IMF 2 ,…,IMF n and a residual term r(t);

[0058] Specifically, the mathematical expression is:

[0059]

[0060] It should be noted that usually the first few intrinsic mode function (IMF) components (high frequency components) contain more noise, while the subsequent intrinsic mode function (IMF) components (low frequency components) contain the main signal information;

[0061] Furthermore, by analyzing the frequency spectrum characteristics or statistical indicators of each eigenmode function component, the noise content in each eigenmode function component is determined;

[0062] Further, a processing method for removing or suppressing noise is selected according to the noise content;

[0063] Specifically, the processing rules are as follows:

[0064] Rule 1: If the noise is to be removed, the intrinsic mode function components that are judged to be noise are directly discarded;

[0065] Rule 2: If the noise is suppressed, threshold processing is applied to the intrinsic mode function components containing some noise to suppress the noise while retaining the key vibration signal;

[0066] Furthermore, the intrinsic mode function components after noise removal or suppression are combined with the residual term to reconstruct the filtered vibration signal;

[0067] Specifically, the filtered vibration signal x denoised (t) The formula is expressed as:

[0068]

[0069] Where k is the starting index of the retained IMF component;

[0070] Furthermore, the vibration information without noise filtering is normalized, and the range of the vibration information without noise filtering after normalization is compressed to between [0, 1];

[0071] Specifically, the normalization formula is expressed as:

[0072]

[0073] in, Represented as the normalized original vibration signal x(t);

[0074] S2, using the MTF-GAF feature image conversion method on the vibration signal data set to generate three feature images converted from vibration signals;

[0075] Furthermore, the characteristic image conversion method of MTF is as follows:

[0076] By discretizing the state of the vibration signal and capturing the transition probability between the states of the vibration signal, the one-dimensional vibration signal is converted into a two-dimensional matrix form;

[0077] Furthermore, the vibration signal is normalized, and the normalized vibration signal is discretized into Q states;

[0078] Specifically, it is divided into Q states, and the signal range is evenly divided into Q intervals, each interval represents a state. Assuming Q = 4, the normalized signal will be divided into four intervals, each interval corresponds to a state {1, 2, 3, 4};

[0079] Furthermore, according to the discretized vibration signal, the transition probability between Q states is calculated to generate a state transfer matrix. By traversing the entire vibration signal, the number of transitions from state i to state j is counted, and the transition probability of each state is calculated to generate a state transfer matrix.

[0080] Specifically, the transition probability between Q states is expressed as:

[0081]

[0082] Among them, 1{·} is an indicator function, which takes the value of 1 when the condition is true, and 0 otherwise; the numerator represents the transition from state s i Transfer to state s j The denominator represents the calculation state s i The number of times it appears in the vibration signal (i.e., how many times from s i transfer initiated);

[0083] Furthermore, the state transfer matrix is ​​used to construct the Markov transition field matrix;

[0084] Specifically, the Markov transition field matrix is ​​expressed as:

[0085] MTF ij =P s(i)，s(j)

[0086] Among them, S(i) and S(j) are the states of the i-th and j-th time points in the vibration signal, respectively;

[0087] It should be noted that since the matrix MTF represents the probability of state transition between different time points, it can reflect the dynamic pattern of the vibration signal;

[0088] Furthermore, the characteristic image conversion method of GASF is as follows:

[0089] A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the cosine value of the angle sum;

[0090] Furthermore, the normalized vibration signal Perform angle mapping to convert each point in the vibration signal into a corresponding angle;

[0091] Specifically, the mapping formula is expressed as:

[0092]

[0093] Where φ(t) is an angle;

[0094] Use the angles obtained by mapping and calculate the cosine value of the angle sum at every two time points to construct the Gram sum matrix;

[0095] Specifically, construct the Gram sum matrix GASF i，j It is expressed as:

[0096] GASF i，j =cos(φ(t i )+φ(t j ))

[0097] Among them, φ(t i ) and φ(t j ) are respectively the time point t i and t j The corresponding angle;

[0098] Furthermore, the feature image conversion method of GADF is as follows:

[0099] A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the angle difference;

[0100] Furthermore, the normalized vibration signal is angle mapped to convert each point in the vibration signal into a corresponding angle;

[0101] Specifically, the mapping formula here is the same as the aforementioned mapping formula;

[0102] Furthermore, the angle obtained by mapping is used to calculate the sine value of the angle difference between every two time points to construct the Gram difference matrix;

[0103] Specifically, construct the Gram difference matrix GADF i，j It is expressed as:

[0104] GADF i，j = sin(φ(t i )-φ(t j ))

[0105] It needs to be explained that each matrix obtained is the final two-dimensional image, that is, the feature image obtained by transformation;

[0106] It should be noted that MTF can capture the dynamic changes of different states in vibration signals and has strong time sequence characteristics; GASF focuses on capturing the global trend information of vibration signals and can intuitively reflect the overall behavior of vibration signals; GADF can capture small fluctuations and mutations in vibration signals and is good at detecting sudden faults (such as cracks and fractures) in vibration signals. The core lies in that it can analyze the information in vibration signals from different angles.

[0107] S3, dividing the converted feature images into a training set and a test set, inputting the feature images of the training set into the VGG-GAM model through multi-channels for training to obtain an optimal model, and then inputting the feature images of the test set into the optimal model for fault diagnosis, outputting the fault type, and forming a fault diagnosis method for the vibration of the stator end of the steam turbine generator;

[0108] Furthermore, based on the VGG network structure, the VGG network architecture consists of multiple VGG convolutional blocks, each of which contains multiple convolutional layers and a maximum pooling layer;

[0109] Furthermore, three feature images generated by the vibration signal transformation are received through the input layer, and the size of the feature images is M×M×1. After normalization, the three feature images are merged into an input tensor with three channels and the size is M×M×3.

[0110] Furthermore, a GAM module is introduced after the convolution layer to extract the global information of the convolution feature map through average pooling and maximum pooling. The pooling results are concatenated and then the attention map is generated through convolution operation and added to the original feature map.

[0111] Furthermore, the feature extraction part of the model consists of multiple VGG convolutional blocks and the introduced GAM module. After the feature extraction is completed, the final feature map is flattened into a one-dimensional vector, passed through two fully connected layers, and the fault category is output using the Softmax classifier;

[0112] Furthermore, the Kaiming method is used to initialize the weights of the convolutional layer;

[0113] Furthermore, the feature image is used as the model input for training. After the model converges, the model is saved (the best model is obtained). The model fault diagnosis effect is verified with the test set. Subsequently, data can be input into the model in real time. If the data is normal, the output is normal. If the data is abnormal, the possible fault type is output, thereby realizing dynamic fault diagnosis.

[0114] Specifically, the architecture code after introducing the GAM module is as follows:

[0115] Block 1:

[0116] Conv2D(64,kernel_size=3x3,padding='same',activation='relu');

[0117] GAM: Global attention mechanism, weighting the convolution output;

[0118] Conv2D(64,kernel_size=3x3,padding='same',activation='relu');

[0119] GAM: Global Attention Mechanism;

[0120] MaxPooling2D(pool_size=2x2, strides=2);

[0121] Block 2:

[0122] Conv2D(128,kernel_size=3x3,padding='same',activation='relu');

[0123] GAM: Global Attention Mechanism;

[0124] Conv2D(128,kernel_size=3x3,padding='same',activation='relu');

[0125] GAM: Global Attention Mechanism;

[0126] MaxPooling2D(pool_size=2x2, strides=2);

[0127] Block 3:

[0128] Conv2D(256,kernel_size=3x3,padding='same',activation='relu');

[0129] GAM: Global Attention Mechanism;

[0130] Conv2D(256,kernel_size=3x3,padding='same',activation='relu');

[0131] GAM: Global Attention Mechanism;

[0132] Conv2D(256,kernel_size=3x3,padding='same',activation='relu');

[0133] GAM: Global Attention Mechanism;

[0134] MaxPooling2D(pool_size=2x2, strides=2);

[0135] Block 4:

[0136] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0137] GAM: Global Attention Mechanism;

[0138] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0139] GAM: Global Attention Mechanism;

[0140] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0141] GAM: Global Attention Mechanism;

[0142] MaxPooling2D(pool_size=2x2, strides=2);

[0143] Block 5:

[0144] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0145] GAM: Global Attention Mechanism;

[0146] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0147] GAM: Global Attention Mechanism;

[0148] Conv2D(512,kernel_size=3x3,padding='same',activation='relu');

[0149] GAM: Global Attention Mechanism;

[0150] MaxPooling2D(pool_size=2x2, strides=2);

[0151] Dense(4096,activation='relu');

[0152] Dropout(0.5);

[0153] Dense(4096,activation='relu');

[0154] Dropout(0.5);

[0155] Dense(number of categories, activation = 'softmax'); Set the output unit according to the number of fault categories required;

[0156] It should be noted that using the Kaiming method to initialize the convolutional layer weights can enable the model to converge faster when starting training.

[0157] Example 2

[0158] Reference Figure 2 , which is the second embodiment of the present invention, provides a fault diagnosis system based on the vibration of the stator end of a steam turbine generator, comprising:

[0159] In the data preprocessing module, referring to the generator vibration guidelines and expert experience, we selected monitoring quantities that can truly and effectively reflect the generator operating status and the types and locations of faults that often occur in the equipment, and created a matching online monitoring system. Under the guidance of the monitoring system, we collected vibration information that best reflects the vibration of the generator stator end, filtered out the noise and abnormal data of the vibration information, and normalized it to create a vibration signal data set.

[0160] In the feature image conversion module, the normalized vibration data transmitted from the data preprocessing module is converted into a feature image. The present invention uses the MTF-GAF feature image conversion method to generate three feature images converted from vibration signals, which can extract effective information from vibration signals from multiple dimensions, better reflect the complexity of the signal, and improve the accuracy and robustness of the fault diagnosis model.

[0161] In the fault diagnosis module, the converted feature images are first divided into training sets and test sets. The training set images are input into the VGG-GAM model through multi-channels for full training to obtain the best model. The test set images are then input into the best model for fault diagnosis, and finally the corresponding fault type is output.

[0162] Those skilled in the art will appreciate that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0163] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program 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 computer program instructions. These computer program instructions can be provided to a processor 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 of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. 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.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing 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.

[0166] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0167] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A fault diagnosis method based on the vibration of the stator end of a steam turbine generator, characterized in that: include: Collecting vibration information of the stator end of the steam turbine generator, filtering out noise of the vibration information and normalizing the vibration information to create a vibration signal data set; The MTF-GAF feature image conversion method is used on the vibration signal data set to generate three feature images converted from the vibration signals; The transformed feature images are divided into a training set and a test set. The feature images of the training set are input into the VGG-GAM model through multi-channels for training to obtain the best model. The feature images of the test set are then input into the best model for fault diagnosis, and the fault type is output to form a fault diagnosis method for the stator end vibration of the steam turbine generator.

2. The fault diagnosis method based on the stator end vibration of the steam turbine generator according to claim 1, characterized in that: The filtering out the noise of the vibration information comprises: The empirical mode decomposition method is used to decompose the original vibration signal into multiple intrinsic mode function components and a residual term; By analyzing the frequency spectrum characteristics or statistical indicators of each intrinsic mode function component, the noise content in each intrinsic mode function component is determined; Selecting a noise removal or noise suppression processing method according to the noise content; The intrinsic mode function component after noise removal or suppression is combined with the residual term to reconstruct the filtered vibration signal.

3. The fault diagnosis method based on the stator end vibration of the steam turbine generator according to claim 2, characterized in that: A method for removing or suppressing noise is selected according to the noise content, including: If the noise is to be removed, the intrinsic mode function components that are judged to be noise are directly discarded; If the processing is noise suppression, threshold processing is applied to the intrinsic mode function components containing some noise to suppress the noise while retaining the key vibration signal.

4. The fault diagnosis method based on the stator end vibration of the steam turbine generator according to claim 1, characterized in that: The vibration information is normalized, including: The vibration information without noise filtered out is normalized, and the range of the vibration information without noise filtered out is compressed to [0, 1].

5. The fault diagnosis method based on the stator end vibration of a steam turbine generator according to claim 2 or 4, characterized in that: The MTF-GAF feature image conversion method is used on the vibration signal data set to generate three feature images converted from vibration signals, including: By discretizing the state of the vibration signal and capturing the transition probability between the states of the vibration signal, the one-dimensional vibration signal is converted into a two-dimensional matrix form; A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the cosine value of the angle sum; A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the angle difference.

6. The fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to claim 5, characterized in that: By discretizing the state of the vibration signal and capturing the transition probability between the states of the vibration signal, the one-dimensional vibration signal information is converted into a two-dimensional matrix form, including: Normalize the vibration signal, and discretize the normalized vibration signal into Q states; According to the discretized vibration signal, the transition probability between Q states is calculated to generate a state transfer matrix. By traversing the entire vibration signal, the number of transitions from state i to state j is counted, and the transition probability of each state is calculated to generate a state transfer matrix. The state transfer matrix is ​​used to construct a Markov transition field matrix.

7. The fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to claim 5, characterized in that: A two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the cosine value of the angle sum, including: Perform angle mapping on the normalized vibration signal, converting each point in the vibration signal into a corresponding angle; The angles obtained by mapping are used to calculate the cosine value of the angle sum of every two time points and construct the Gram sum matrix.

8. The fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to claim 5, characterized in that: The two-dimensional image is generated by mapping the one-dimensional vibration signal to the angle space and calculating the angle difference, including: Perform angle mapping on the normalized vibration signal, converting each point in the vibration signal into a corresponding angle; The angle obtained by mapping is used to calculate the sine value of the angle difference between every two time points to construct the Gram difference matrix.

9. The fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to claim 1 or 5, characterized in that: VGG-GAM model, including: Based on the VGG network structure, the VGG network architecture consists of multiple VGG convolution blocks, each of which includes multiple convolution layers and a maximum pooling layer; The input layer receives three feature images generated by the vibration signal transformation, wherein the size of the feature images is M×M×1, and the three feature images are normalized and merged into an input tensor with three channels, wherein the size is M×M×3; The GAM module is introduced after the convolution layer to extract the global information of the convolution feature map through average pooling and maximum pooling. The pooling results are concatenated and then the attention map is generated through convolution operation and added to the original feature map. The feature extraction part of the model consists of multiple VGG convolution blocks and the introduced GAM module. After the feature extraction is completed, the final feature map is flattened into a one-dimensional vector, passed through two fully connected layers, and the fault category is output using the Softmax classifier.

10. The fault diagnosis method based on the vibration of the stator end of the steam turbine generator according to claim 9, characterized in that: Also includes: The Kaiming method is used to initialize the convolutional layer weights.

Images