Fault diagnosis method, system and related device for rotating equipment of nuclear power plant

Abstract

The invention discloses a fault diagnosis method and system for nuclear power plant rotating equipment and a related device, and relates to the field of nuclear power plants, and the method comprises the steps: obtaining a vibration signal sequence composed of a plurality of target vibration signals of preset duration from collected vibration signals, and converting each target vibration signal into time domain data and frequency domain data; for each target vibration signal, time-frequency features and frequency-domain features are extracted by using a first residual network containing convolution kernels with different scales, and the time-frequency-domain features are obtained through splicing; contextual information of the time-frequency domain features is obtained through a self-attention mechanism, and overall data features of the vibration signals are obtained through combination; and classifying the overall data features, and taking a classification result of the overall data features as a fault type corresponding to the vibration signal. According to the invention, the features of the vibration signals are comprehensively captured from the frequency domain and the time domain through the first residual network, and classification is carried out after the context information of the time-frequency domain features is synthesized through a self-attention mechanism, so that the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] The present application relates to the technical field of nuclear power plants, and in particular to a fault diagnosis method, system and related devices for rotating equipment in a nuclear power plant. Background Art

[0002] In nuclear power plants, rotating equipment, such as motors, is crucial for normal operation and efficient power generation. To maintain the continuity of nuclear power production, downtime for maintenance and repairs should be minimized throughout the equipment's lifecycle. Therefore, fault diagnosis of rotating equipment plays a crucial role in nuclear power production. Rapidly identifying the fault type through fault diagnosis and reducing downtime for maintenance is currently a major research area. Summary of the Invention

[0003] In view of the above problems, this application provides a fault diagnosis method, system and related devices for rotating equipment in nuclear power plants to achieve the purpose of improving the fault diagnosis accuracy of rotating equipment. The specific solution is as follows:

[0004] A first aspect of the present application provides a method for diagnosing faults of rotating equipment in a nuclear power plant, the method comprising:

[0005] Obtaining a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and converting each of the target vibration signals into time domain data and frequency domain data respectively;

[0006] For each target vibration signal, a first residual network is used to extract the time-frequency features of the time domain data and the frequency domain features of the frequency domain data, wherein the first residual network includes multiple convolution kernels, each of which has a different scale. The first residual network includes a main path and a jump connection path, and the output of the main path is connected to the output of the jump connection path. The main path includes a first branch and a second branch. The first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; the second branch includes a batch normalization layer, an activation function, and a convolution kernel of one scale. The output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path.

[0007] splicing the time-frequency features and frequency domain features of each target vibration signal to obtain the time-frequency domain features of each target vibration signal;

[0008] Acquire the context information of each of the time-frequency domain features through a self-attention mechanism;

[0009] combining a plurality of time-frequency domain features with the context information to obtain an overall data feature of the vibration signal;

[0010] Classifying the overall data features, and using the classification results of the overall data features as the fault type corresponding to the vibration signal;

[0011] The extracting the time-frequency features of the time-domain data and the frequency-domain features of the frequency-domain data using the first residual network includes:

[0012] For the time domain data, extract a first time-frequency feature of the time domain data through the first branch, extract a second time-frequency feature of the time domain data through the second branch, and after splicing the first time-frequency feature and the second time-frequency feature at the splicing layer, superimpose them with the time domain data of the skip connection path to obtain the time-frequency feature;

[0013] For the frequency domain data, the first frequency domain feature of the frequency domain data is extracted through the first branch, and the second frequency domain feature of the frequency domain data is extracted through the second branch. After the first frequency domain feature and the second frequency domain feature are spliced in the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain feature.

[0014] In one possible implementation, obtaining the context information of each of the time-frequency domain features through the self-attention mechanism includes:

[0015] For a target time-frequency domain feature, respectively calculating a similarity score between each other time-frequency domain feature and the target time-frequency domain feature, and converting the similarity score into an attention weight, where the target time-frequency domain feature is one of the multiple time-frequency domain features;

[0016] Obtaining multiple attention weights of the target time-frequency domain feature as context information of the target time-frequency domain feature, each of the attention weights corresponding to one of the other time-frequency domain features;

[0017] Context information of each of the time-frequency domain features is obtained.

[0018] In a possible implementation, combining multiple time-frequency domain features with the context information to obtain the overall data features of the vibration signal includes:

[0019] A plurality of time-frequency domain features with the context information are spliced together through a fully connected layer, and the spliced features are used as the overall data features of the vibration signal.

[0020] In a possible implementation, the method further includes:

[0021] Acceleration vibration data of the rotating equipment is collected as the vibration signal.

[0022] A second aspect of the present application provides a fault diagnosis system for rotating equipment in a nuclear power plant, the fault diagnosis system for rotating equipment in a nuclear power plant comprising:

[0023] a conversion unit, configured to obtain a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and convert each of the target vibration signals into time domain data and frequency domain data;

[0024] An extraction unit is configured to extract, for each target vibration signal, the time-frequency features of the time domain data and the frequency domain features of the frequency domain data using a first residual network, wherein the first residual network includes multiple convolution kernels, each convolution kernel having a different scale, and the first residual network includes a main path and a skip connection path, wherein the output of the main path is connected to the output of the skip connection path, and the main path includes a first branch and a second branch, wherein the first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; and the second branch includes a batch normalization layer, an activation function, and a convolution kernel of one scale; the output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path;

[0025] a splicing unit, configured to splice the time-frequency features and frequency domain features of each target vibration signal to obtain the time-frequency domain features of each target vibration signal;

[0026] An acquisition unit, configured to acquire context information of each of the time-frequency domain features through a self-attention mechanism;

[0027] a combining unit, configured to combine a plurality of time-frequency domain features with the context information to obtain an overall data feature of the vibration signal;

[0028] a classification unit, configured to classify the overall data features and use the classification result of the overall data features as the fault type corresponding to the vibration signal,

[0029] The extraction unit is specifically configured as follows:

[0030] For time domain data, the first time-frequency feature of the time domain data is extracted through the first branch, and the second time-frequency feature of the time domain data is extracted through the second branch. The first time-frequency feature and the second time-frequency feature are concatenated at the concatenation layer and then superimposed with the time domain data of the jump connection path to obtain the time-frequency feature;

[0031] For frequency domain data, the first frequency domain features of the frequency domain data are extracted through the first branch, and the second frequency domain features of the frequency domain data are extracted through the second branch. After the first frequency domain features and the second frequency domain features are spliced at the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain features.

[0032] In a possible implementation, the first residual network includes a main path and a skip connection path, the main path includes a first branch and a second branch, and the first residual network is specifically configured as follows:

[0033] The output of the main path is connected to the output of the skip connection path;

[0034] The first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales.

[0035] The second branch includes a batch normalization layer, an activation function, and a convolution kernel of a certain scale;

[0036] The output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path.

[0037] A third aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0038] The memory is used to store computer programs;

[0039] The processor is used to execute the computer program so that the electronic device can implement the fault diagnosis method for rotating equipment in a nuclear power plant according to the first aspect or any implementation method of the first aspect.

[0040] The fourth aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the fault diagnosis method for rotating equipment in a nuclear power plant according to the first aspect or any implementation method of the first aspect.

[0041] By means of the above technical solution, the present application provides a fault diagnosis method, system and related devices for rotating equipment in nuclear power plants. Since the vibration signal of the rotating equipment is a time-varying signal, the present method converts the vibration signal into time domain data and frequency domain data, and uses the first residual network to extract the time-frequency features of the time domain data and the frequency domain features of the frequency domain data. Since the first residual network is provided with multiple convolution kernels of different scales, the features of multiple receptive fields can be extracted from the time domain data and the frequency domain data to fully capture the features of the vibration signal, and the time-frequency features and the frequency domain features can be integrated through feature splicing. Since both the time-frequency features and the frequency domain features are localized features, the contextual information of the time-frequency domain features is obtained through the self-attention mechanism and the time-frequency domain features are combined to obtain the overall data features of the vibration signal, and then the overall data features are classified to determine the fault type corresponding to the vibration data. This method adapts to the characteristics of vibration signals from rotating equipment, analyzing them not only in the frequency domain but also in the time domain. It uses a first residual network with multiple convolution kernels of different scales for feature extraction to comprehensively capture the characteristics of the vibration signal. A self-attention mechanism integrates the contextual information of the time and frequency domain features to obtain the overall data characteristics of the vibration signal before classification, thereby improving classification accuracy. Therefore, this method can effectively improve the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0043] Figure 1 A flowchart of a method for diagnosing faults of rotating equipment in a nuclear power plant provided in an embodiment of the present application;

[0044] Figure 2 A schematic diagram of the structure of a first residual network provided in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of sample point distribution of training sample data provided in an embodiment of the present application;

[0046] Figure 4 A schematic diagram of sample point distribution of a first residual network output provided in an embodiment of the present application;

[0047] Figure 5 A schematic diagram of the distribution of input sample points without a self-attention mechanism provided in an embodiment of the present application;

[0048] Figure 6A schematic diagram of the distribution of input sample points with a self-attention mechanism provided in an embodiment of the present application;

[0049] Figure 7 A schematic diagram of the structure of a fault diagnosis system for rotating equipment in a nuclear power plant provided in an embodiment of the present application;

[0050] Figure 8 This is a hardware structure block diagram of an electronic device provided in this application. DETAILED DESCRIPTION

[0051] The following describes the embodiments of the present application in conjunction with the accompanying drawings. The terms used in the implementation methods of the present application are only used to explain the specific embodiments of the present application and are not intended to limit the present application.

[0052] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0053] The terms "first", "second" etc. in the specification of the application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, and this is merely a way of distinguishing the objects of the same attributes when describing them in the embodiments of the application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0054] Rotating equipment such as motors plays an important role in the production of various industries. In the actual operation of nuclear power plants, rotating equipment such as motors can be used for the normal operation of nuclear power plants and efficient power generation. However, when rotating equipment fails, it will hinder the normal production of the nuclear power plant, and serious equipment failures may cause the rotating equipment to be scrapped, thereby causing nuclear safety production accidents. Therefore, in order to maintain the continuity of nuclear power production, the downtime and maintenance time of rotating equipment when it fails should be reduced as much as possible during its life cycle to reduce the impact on the operation of the nuclear power plant. Fault diagnosis of rotating equipment plays an important role in the nuclear power production of nuclear power plants. Through fault diagnosis, the fault type of rotating equipment can be quickly determined to reduce the downtime and maintenance time of rotating equipment. It is currently the main research direction of fault diagnosis.

[0055] Fault diagnosis for rotating equipment can be broadly categorized into two approaches: those based on the fault mechanism and those based on artificial intelligence algorithms. These approaches, in turn, can be further categorized into those based on statistical pattern recognition and those based on artificial neural networks. Deep learning networks, a type of artificial neural network, have been widely used in fault diagnosis for rotating equipment. These networks acquire vibration signals, extract frequency-domain features from the signal's spectrum using a single-scale convolution kernel, and then perform feature classification to determine the fault type.

[0056] Common fault types of rotating equipment may include cavitation, loose base, rotor imbalance, rotor misalignment, and failures of major components such as bearings. The vibration spectra of some fault types are very similar. The deep learning network model only extracts frequency domain features through a single-scale convolution kernel, making it difficult to fully capture the characteristics of the vibration signal, resulting in low fault diagnosis accuracy.

[0057] In order to solve the above problems, an embodiment of the present application provides a fault diagnosis method for rotating equipment in a nuclear power plant. This method analyzes the fault information of rotating equipment from the time domain and frequency domain perspectives of the vibration signal, and performs feature extraction processing through multiple convolution kernels of different sizes. It also incorporates a self-attention mechanism to obtain overall data features with contextual information and classify them, effectively improving the accuracy of fault diagnosis and facilitating early intervention in rotating equipment to avoid the expansion of faults and prolonging the downtime for maintenance or leading to more serious consequences. The following is a detailed introduction to the fault diagnosis method for rotating equipment in a nuclear power plant according to an embodiment of the present application, combined with the accompanying drawings.

[0058] Reference Figure 1 , Figure 1 A flowchart of a method for diagnosing a fault of rotating equipment in a nuclear power plant provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, a fault diagnosis method for rotating equipment in a nuclear power plant provided by an embodiment of the present application may include steps S10 to S15, and these steps are described in detail below.

[0059] S10. Obtain a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and convert each target vibration signal into time domain data and frequency domain data respectively.

[0060] Among them, the vibration signal can be the vibration data of the rotating equipment during operation, and can include information such as vibration amplitude. Since rotating equipment is often accompanied by abnormal vibration and noise when it fails, and the amplitude of vibration and other information are not exactly the same for different types of failures, when the rotating equipment fails, this embodiment can collect the vibration signal of the rotating equipment for fault diagnosis. Specifically, this embodiment can collect the acceleration vibration data of the rotating equipment through a wireless vibration sensor installed on the rotating equipment, and use the acceleration vibration data as a vibration signal. The wireless vibration sensor can include an acceleration sensor and a velocity sensor. The wireless vibration sensor can be used to capture the acceleration changes in the vibration structure of the rotating equipment and convert it into an electrical signal for recording and analysis. The acceleration vibration data can reflect the vibration intensity and impact force of the rotating equipment.

[0061] The target vibration signal may be a vibration signal of a preset duration in the collected vibration signal. Specifically, the preset duration may be 1 second, 5 seconds, etc., and may be determined according to actual conditions.

[0062] Time-domain data can be data on the relationship between the vibration signal and time, and can directly reflect how the vibration signal changes over time. In this embodiment, the time-domain data can be a time-domain graph, where time can be the horizontal axis and the amplitude or instantaneous value of the vibration signal can be the vertical axis, reflecting how the vibration signal changes over time. Frequency-domain data can be data on the frequency characteristics of the vibration signal. Frequency-domain data can be obtained by converting the vibration signal through mathematical transformations (such as Fourier transforms). Frequency-domain data can reflect the amplitude and phase information of the vibration signal at different frequencies, reflecting the intrinsic frequency characteristics of the vibration signal. In this embodiment, the frequency-domain data can be a spectrogram, where frequency can be the horizontal axis and the amplitude or power of the signal can be the vertical axis.

[0063] S11. For each target vibration signal, extract time-frequency features of time-domain data and frequency-domain features of frequency-domain data using a first residual network, where the first residual network includes multiple convolution kernels, each of which has a different scale;

[0064] S12. Concatenate the time-frequency features and frequency-domain features of each target vibration signal to obtain the time-frequency-domain features of each target vibration signal.

[0065] The residual network is a deep neural network structure composed of multiple residual blocks. Each residual block contains a convolutional layer and a skip connection, which allows the input signal to be directly superimposed on the output without going through the convolutional layer.

[0066] To extract more comprehensive features of vibration signals, this embodiment improves the residual network structure, resulting in the first residual network of this embodiment. The first residual network of this embodiment may include a main path (convolutional layer) and a skip connection path. The output of the main path is connected to the output of the skip connection path. The main path includes a first branch and a second branch. The first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; the second branch includes a batch normalization layer, an activation function, and a convolution kernel of one scale. The output of the first branch and the output of the second branch are connected via a concatenation layer, and the concatenated output serves as the output of the main path. The convolution kernel is used to extract features from time domain data and frequency domain data, respectively. The batch normalization (BN) layer normalizes the output of the convolution kernel to a mean of 0 and a variance of 1, ensuring that the input to the activation function conforms to a standard normal distribution. The activation function introduces nonlinearity to improve the expressive power of the first residual network. Specifically, the activation function sets all negative values to 0 and leaves all positive values unchanged. In the residual function, ReLU (Rectified Linear Unit) is often used as the activation function. Figure 2 As shown, the output of the main path is connected to the output of the skip connection path to superimpose the output of the main path and the output of the skip connection path. The main path includes a first branch and a second branch. The first branch can include a one-dimensional convolution (Conv1d(9,2)) with a scale of 9×2 and a one-dimensional convolution kernel (Conv1d(3,1)) with a scale of 3×1, and a batch normalization layer (BN) and an activation function (ReLU) are connected after each convolution kernel. The second branch includes a one-dimensional convolution kernel (Conv1d(3,1)) with a scale of 3×1, and a batch normalization layer (BN) and an activation function (ReLU) are connected after the convolution kernel. The output of the first branch and the output of the second branch are connected through a concatenation layer (Concat), and the output of the concatenation layer is used as the output of the main path.

[0067] Based on the first residual network structure described above, the specific process of extracting the time-frequency features of time-domain data and the frequency-domain features of frequency-domain data in this embodiment can be as follows:

[0068] For time domain data, the first branch extracts the first time-frequency features of the time domain data, and the second branch extracts the second time-frequency features of the time domain data. After splicing the first and second time-frequency features at the splicing layer, they are superimposed with the time domain data of the jump connection path to obtain the time-frequency features. For frequency domain data, the first branch extracts the first frequency domain features of the frequency domain data, and the second branch extracts the second frequency domain features of the frequency domain data. After splicing the first and second frequency domain features at the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain features. Finally, the time-frequency features and frequency domain features are spliced together to obtain the time-frequency domain features of the target vibration signal.

[0069] Of course, in another optional embodiment, the time-frequency features of the time domain data can be extracted through the first branch, the frequency domain features of the frequency domain data can be extracted through the second branch, and the time-frequency features and the frequency domain features can be directly spliced through the splicing layer to obtain the time-frequency domain features of the target vibration signal.

[0070] This embodiment converts the vibration signal into time domain data and frequency domain data, which can be used to analyze the vibration signal from multiple perspectives. Moreover, by using convolution kernels of different scales, features of different resolutions (different receptive fields) can be extracted to obtain more comprehensive data features from the vibration signal.

[0071] S13, obtain the context information of each time-frequency domain feature through the self-attention mechanism;

[0072] S14, combining multiple time-frequency domain features with context information to obtain overall data features of the vibration signal;

[0073] S15. Classify the overall data features, and use the classification results of the overall data features as the fault type corresponding to the vibration signal.

[0074] Among them, the Self-Attention Mechanism captures long-distance dependencies by dynamically calculating the correlation between different positions in the sequence, and gives the model the ability to globally understand the input sequence.

[0075] In this embodiment, the self-attention structure unit of the non-local block is used to obtain the context information of the time-frequency domain features. Specifically, the time-frequency domain feature sequence of the vibration signal is converted into the input matrix (Each row corresponds to a vector representation of time-frequency domain features), and then calculate the input matrix The corresponding query matrix (Each row can correspond to a query vector representation of time-frequency domain features), key matrix (Each row can correspond to a key vector representation of a time-frequency domain feature) and a value matrix (Each row can correspond to a value vector representation of a time-frequency domain feature), the specific calculation formula can be shown as follows:

[0076] ;

[0077] ;

[0078] ;

[0079] in, It can represent the weight matrix corresponding to the query vector; It can represent the weight matrix corresponding to the key vector; It can represent the weight matrix corresponding to the value vector.

[0080] After obtaining the query vector, key vector, and value vector of each time-frequency domain feature, the subsequent steps are described using the target time-frequency domain feature as an example. The target time-frequency domain feature is one of the multiple time-frequency domain features.

[0081] Specifically, for the target time-frequency domain feature, the similarity score between each other time-frequency domain feature and the target time-frequency domain feature is calculated respectively. The calculation formula for the similarity score between one other time-frequency domain feature and the target time-frequency domain feature can be as follows:

[0082] ;

[0083] in, It can represent the similarity score between other time-frequency domain features and target time-frequency domain features. From the perspective of vector, Represents a vector and vector The inner product of A query vector that can represent the target time-frequency domain features; It can represent the key vector of other time-frequency domain features. Therefore, for the target time-frequency domain feature, the inner product of the query vector of the target time-frequency domain feature and the key vector of each other time-frequency domain feature is calculated as the similarity score between the target time-frequency domain feature and each other time-frequency domain feature.

[0084] After obtaining the similarity score between the target time-frequency domain feature and each other time-frequency domain feature, the similarity score is converted into an attention weight, and multiple attention weights of the target time-frequency domain feature are obtained. The multiple attention weights of the target time-frequency domain feature are used as the context information of the target time-frequency domain feature, and each attention weight corresponds to another time-frequency domain feature, thereby obtaining the context information of each time-frequency domain feature. The attention weight can represent the degree of correlation between two time-frequency domain features.

[0085] Specifically, the calculation formula for the attention weight between another time-frequency domain feature and the target time-frequency domain feature can be as follows:

[0086] ;

[0087] in, It can represent the attention weight between another time-frequency domain feature and the target time-frequency domain feature, that is, The key vector of the time-frequency domain feature is The attention weight of the query vector of time-frequency domain features; Can represent exponential functions; The number of other time-frequency domain features that can be represented.

[0088] After obtaining multiple attention weights of the target time-frequency domain features, the vector representation of the target time-frequency domain features can be obtained by weighted summing of multiple attention weights. That is, the sum of the weighted sums of multiple attention weights represents the target time-frequency domain features. The vector representation of the target time-frequency domain features can be shown as follows:

[0089] ;

[0090] in, It can be a vector representation of the target time-frequency domain features; It can represent the attention weight between another time-frequency domain feature and the target time-frequency domain feature; It can represent the value vector of other time-frequency domain features. From the perspective of vectors, Can be the first Vector representation of time-frequency domain features; Can indicate the The key vector of the time-frequency domain feature is The attention weight of the query vector of time-frequency domain features; Can indicate the The value vector of time-frequency domain features.

[0091] Therefore, this embodiment obtains the context information of each time-frequency domain feature through the self-attention mechanism, that is, each time-frequency domain feature is converted into a vector representation, and before classification, multiple time-frequency domain features with context information are spliced through a fully connected layer, and the spliced features are used as the overall data features of the vibration signal to reflect the overall data characteristics of the vibration signal. After this embodiment obtains the overall data features of the vibration signal, the overall data features can be classified to determine the fault type of the fault occurring in the rotating equipment, so as to facilitate timely and accurate intervention. Specifically, this embodiment can achieve classification through the SoftMax layer.

[0092] Furthermore, in the actual experimental process of this embodiment, the fault diagnosis model in this embodiment is divided into three parts: the backbone network (the first residual network in this embodiment), the self-attention mechanism and the classifier, and the training sample data (which may include data of normal working conditions and various fault working conditions) is mapped to a two-dimensional space, and the categories are distinguished by different color markings, such as Figure 3 As shown, each color can represent a type of working condition, each point can represent a sample point, and each type of sample points is randomly scattered.

[0093] This embodiment can be determined through experiments that the first residual network can effectively improve the classification accuracy. Specifically, various types of training sample data are input into the fault diagnosis model. After being processed by the backbone network, the output sample point distribution can be as follows: Figure 4 By comparison Figure 3 and Figure 4 The highly random training sample data, when passed through the first residual network of this embodiment, can concentrate sample points for similar operating conditions. Since vibration signals are time-varying vibration data whose time-frequency domain features change rapidly over time, they exhibit strong randomness. Therefore, based on the above results, it can be determined that the first residual network of this embodiment can better extract the time-frequency domain features of different fault conditions from vibration signals, helping to improve subsequent classification accuracy.

[0094] Furthermore, this embodiment can also confirm through experiments that the self-attention mechanism can effectively improve the classification accuracy. Figure 5 and Figure 6 As shown, Figure 5 This is a schematic diagram of the input sample point distribution of the classifier without the self-attention mechanism. Figure 6 This is a diagram showing the distribution of input sample points of the classifier when it has a self-attention mechanism. Figure 5 and Figure 6 It can be seen that with the self-attention mechanism, sample points of the same type are more concentrated, especially green sample points. Therefore, introducing the self-attention mechanism can make the feature distribution of sample data of similar working conditions more concentrated, helping the classifier to better classify. Furthermore, this embodiment also compares the fault diagnosis results of the fault diagnosis model with the self-attention mechanism under different signal-to-noise ratios. The fault diagnosis model with the self-attention mechanism generally has higher classification accuracy than the fault diagnosis model without the self-attention mechanism.

[0095] The embodiment of the present application provides a fault diagnosis method for rotating equipment in a nuclear power plant. Since the vibration signal of the rotating equipment is a time-varying signal, the method converts the vibration signal into time domain data and frequency domain data, and uses a first residual network to extract the time-frequency features of the time domain data and the frequency domain features of the frequency domain data respectively. Since the first residual network is provided with multiple convolution kernels of different scales, it is possible to extract features of multiple receptive fields from the time domain data and the frequency domain data to fully capture the features of the vibration signal, and to integrate the time-frequency features and the frequency domain features through feature splicing. Since both the time-frequency features and the frequency domain features are localized features, the contextual information of the time-frequency domain features is obtained through a self-attention mechanism and the time-frequency domain features are combined to obtain the overall data features of the vibration signal. The overall data features are then classified to determine the fault type corresponding to the vibration data. This method adapts to the characteristics of vibration signals from rotating equipment, analyzing them not only in the frequency domain but also in the time domain. It uses a first residual network with multiple convolution kernels of different scales for feature extraction to comprehensively capture the characteristics of the vibration signal. A self-attention mechanism integrates the contextual information of the time and frequency domain features to obtain the overall data characteristics of the vibration signal before classification. This optimizes the distribution of similar data to improve classification accuracy. Therefore, this method can effectively improve the accuracy of fault diagnosis.

[0096] The above describes a fault diagnosis method for rotating equipment in a nuclear power plant provided by an embodiment of the present application. The following describes a system that applies the above fault diagnosis method for rotating equipment in a nuclear power plant.

[0097] See also Figure 7 , Figure 7 This is a schematic diagram of a fault diagnosis system for rotating equipment in a nuclear power plant provided in an embodiment of the present application. Figure 7 As shown, the fault diagnosis system for rotating equipment in a nuclear power plant may include:

[0098] The conversion unit 100 is configured to obtain a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and convert each target vibration signal into time domain data and frequency domain data respectively;

[0099] An extraction unit 110 is configured to extract, for each target vibration signal, time-frequency features of the time domain data and frequency-domain features of the frequency domain data using a first residual network, wherein the first residual network includes multiple convolution kernels, each of which has a different scale;

[0100] A splicing unit 120 is used to splice the time-frequency features and frequency domain features of each target vibration signal to obtain the time-frequency domain features of each target vibration signal;

[0101] an acquisition unit 130 for acquiring contextual information of each time-frequency domain feature through a self-attention mechanism;

[0102] A combining unit 140 is configured to combine multiple time-frequency domain features with context information to obtain overall data features of the vibration signal;

[0103] The classification unit 150 is used to classify the overall data features and use the classification results of the overall data features as the fault type corresponding to the vibration signal.

[0104] In one possible implementation, the first residual network includes a main path and a skip connection path, the main path includes a first branch and a second branch, and the first residual network is specifically configured as follows:

[0105] The output of the main path is connected to the output of the skip connection path;

[0106] In the first branch, it contains a batch normalization layer, an activation function, and two convolution kernels of different scales;

[0107] In the second branch, it contains a batch normalization layer, an activation function, and a convolution kernel of a certain scale;

[0108] The output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path.

[0109] In a possible implementation, the extraction unit 110 may be specifically configured as follows:

[0110] For time domain data, the first time-frequency feature of the time domain data is extracted through the first branch, and the second time-frequency feature of the time domain data is extracted through the second branch. The first time-frequency feature and the second time-frequency feature are concatenated at the concatenation layer and then superimposed with the time domain data of the jump connection path to obtain the time-frequency feature;

[0111] For frequency domain data, the first frequency domain features of the frequency domain data are extracted through the first branch, and the second frequency domain features of the frequency domain data are extracted through the second branch. After the first frequency domain features and the second frequency domain features are spliced at the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain features.

[0112] In a possible implementation, the acquiring unit 130 may be specifically configured as follows:

[0113] For the target time-frequency domain feature, the similarity scores between each other time-frequency domain feature and the target time-frequency domain feature are calculated respectively, and the similarity scores are converted into attention weights. The target time-frequency domain feature is one of the multiple time-frequency domain features.

[0114] Obtain multiple attention weights of the target time-frequency domain features as context information of the target time-frequency domain features, where each attention weight corresponds to another time-frequency domain feature;

[0115] Obtain context information for each time-frequency domain feature.

[0116] In a possible implementation, the combining unit 140 may be specifically configured as follows:

[0117] Multiple time-frequency domain features with contextual information are spliced through a fully connected layer, and the spliced features are used as the overall data features of the vibration signal.

[0118] In a possible implementation, the fault diagnosis system for rotating equipment in a nuclear power plant may further include a collection unit:

[0119] The acquisition unit is used to collect acceleration vibration data of rotating equipment as vibration signals.

[0120] An electronic device is also provided in an embodiment of the present application. Figure 8 , which shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application may include but is not limited to fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 8 The electronic device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0121] like Figure 8 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 801, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 802 or programs loaded from a storage device 808 into a random access memory (RAM) 803. When the electronic device is powered on, the RAM 803 also stores various programs and data required for the operation of the electronic device. The processing device 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0122] Typically, the following devices may be connected to the I / O interface 805: an input device 806 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 808 including, for example, a memory card, a hard disk, etc.; and a communication device 809. The communication device 809 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Figure 8 The electronic device is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0123] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the fault diagnosis methods for rotating equipment in a nuclear power plant provided in the embodiment of the present application.

[0124] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When one or more computer programs are executed by an electronic device, the electronic device can implement any fault diagnosis method for rotating equipment in a nuclear power plant provided in an embodiment of the present application.

[0125] It should also be noted that the system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the system embodiments provided in this application, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines.

[0126] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course can also be implemented by special hardware including application-specific integrated circuits, special CPUs, special memories, special components, etc. In general, all functions performed by computer programs can be easily implemented with corresponding hardware, and the specific hardware structures used to implement the same function can also be diverse, such as analog circuits, digital circuits or special circuits, etc. However, for the present application, software program implementation is a better implementation method in most cases. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., and includes a number of instructions to enable a computer device (which can be a personal computer, training equipment, or network equipment, etc.) to execute the methods described in each embodiment of the present application.

[0127] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0128] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, a computer, a training device or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website, a computer, a training device or a data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0129] Each embodiment in this specification is described in a related manner. Similar portions between the embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. In particular, the device embodiments are generally similar to the method embodiments, so their description is relatively simple. For related portions, refer to the description of the method embodiments.

[0130] It is understandable that before using the technical solutions disclosed in the various embodiments of this disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0131] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included in the protection scope of the present application.

Claims

1. A method for fault diagnosis of rotating equipment in a nuclear power plant, characterized in that: The fault diagnosis method for rotating equipment in a nuclear power plant comprises: Obtaining a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and converting each of the target vibration signals into time domain data and frequency domain data respectively; For each target vibration signal, a first residual network is used to extract the time-frequency features of the time domain data and the frequency domain features of the frequency domain data, wherein the first residual network includes multiple convolution kernels, each of which has a different scale. The first residual network includes a main path and a jump connection path, and the output of the main path is connected to the output of the jump connection path. The main path includes a first branch and a second branch. The first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; the second branch includes a batch normalization layer, an activation function, and a convolution kernel of one scale. The output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path. splicing the time-frequency features and frequency domain features of each target vibration signal to obtain the time-frequency domain features of each target vibration signal; Acquire the context information of each of the time-frequency domain features through a self-attention mechanism; combining a plurality of time-frequency domain features with the context information to obtain an overall data feature of the vibration signal; Classifying the overall data features, and using the classification results of the overall data features as the fault type corresponding to the vibration signal; The extracting the time-frequency features of the time-domain data and the frequency-domain features of the frequency-domain data using the first residual network includes: For the time domain data, extract a first time-frequency feature of the time domain data through the first branch, extract a second time-frequency feature of the time domain data through the second branch, and after splicing the first time-frequency feature and the second time-frequency feature at the splicing layer, superimpose them with the time domain data of the skip connection path to obtain the time-frequency feature; For the frequency domain data, the first frequency domain feature of the frequency domain data is extracted through the first branch, and the second frequency domain feature of the frequency domain data is extracted through the second branch. After the first frequency domain feature and the second frequency domain feature are spliced in the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain feature.

2. The fault diagnosis method for rotating equipment in a nuclear power plant according to claim 1, characterized in that: The obtaining of context information of each of the time-frequency domain features through the self-attention mechanism includes: For a target time-frequency domain feature, respectively calculating a similarity score between each other time-frequency domain feature and the target time-frequency domain feature, and converting the similarity score into an attention weight, where the target time-frequency domain feature is one of the multiple time-frequency domain features; Obtaining multiple attention weights of the target time-frequency domain feature as context information of the target time-frequency domain feature, each of the attention weights corresponding to one of the other time-frequency domain features; Context information of each of the time-frequency domain features is obtained.

3. The fault diagnosis method for rotating equipment in a nuclear power plant according to claim 1, characterized in that: Combining the multiple time-frequency domain features with the context information to obtain the overall data features of the vibration signal includes: A plurality of time-frequency domain features with the context information are spliced together through a fully connected layer, and the spliced features are used as the overall data features of the vibration signal.

4. The fault diagnosis method for rotating equipment in a nuclear power plant according to claim 1, characterized in that: Also includes: Acceleration vibration data of the rotating equipment is collected as the vibration signal.

5. A fault diagnosis system for rotating equipment in a nuclear power plant, characterized in that: The fault diagnosis system for rotating equipment in a nuclear power plant includes: a conversion unit, configured to obtain a vibration signal sequence consisting of a plurality of target vibration signals of preset duration from the collected vibration signal, and convert each of the target vibration signals into time domain data and frequency domain data; An extraction unit is configured to extract, for each target vibration signal, the time-frequency features of the time domain data and the frequency domain features of the frequency domain data using a first residual network, wherein the first residual network includes multiple convolution kernels, each convolution kernel having a different scale, and the first residual network includes a main path and a skip connection path, wherein the output of the main path is connected to the output of the skip connection path, and the main path includes a first branch and a second branch, wherein the first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; and the second branch includes a batch normalization layer, an activation function, and a convolution kernel of one scale; the output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path; a splicing unit, configured to splice the time-frequency features and frequency domain features of each target vibration signal to obtain the time-frequency domain features of each target vibration signal; An acquisition unit, configured to acquire context information of each of the time-frequency domain features through a self-attention mechanism; a combining unit, configured to combine a plurality of time-frequency domain features with the context information to obtain an overall data feature of the vibration signal; a classification unit, configured to classify the overall data features and use the classification result of the overall data features as the fault type corresponding to the vibration signal, The extraction unit is specifically configured as follows: For time domain data, the first time-frequency feature of the time domain data is extracted through the first branch, and the second time-frequency feature of the time domain data is extracted through the second branch. The first time-frequency feature and the second time-frequency feature are concatenated at the concatenation layer and then superimposed with the time domain data of the jump connection path to obtain the time-frequency feature; For frequency domain data, the first frequency domain features of the frequency domain data are extracted through the first branch, and the second frequency domain features of the frequency domain data are extracted through the second branch. After the first frequency domain features and the second frequency domain features are spliced at the splicing layer, they are superimposed with the frequency domain data of the jump connection path to obtain the frequency domain features.

6. The fault diagnosis system for rotating equipment in a nuclear power plant according to claim 7, characterized in that: The first residual network includes a main path and a skip connection path, the main path includes a first branch and a second branch, and the first residual network is specifically configured as follows: The output of the main path is connected to the output of the skip connection path; The first branch includes a batch normalization layer, an activation function, and two convolution kernels of different scales; The second branch includes a batch normalization layer, an activation function, and a convolution kernel of a certain scale; The output of the first branch and the output of the second branch are connected through a splicing layer, and the spliced output is used as the output of the main path.

7. An electronic device, characterized in that: comprising at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so that the electronic device can implement the fault diagnosis method for rotating equipment in a nuclear power plant as claimed in any one of claims 1 to 4.

8. A computer program product, characterized in that The method comprises computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the fault diagnosis method for rotating equipment in a nuclear power plant as claimed in any one of claims 1 to 4.

Images