Anti-noise fault classification method and device, equipment, storage medium and product
Through signal decomposition and feature extraction of nuclear steam turbine vibration signals, combined with fault classification sub-models, the problem of difficulty in capturing noise signals with existing technology is solved, and accurate identification of faults that are difficult to distinguish is achieved.
Patent Information
- Application Number
- CN202411343409.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2025-01-21
AI Technical Summary
The prior art is difficult to clearly capture the noise signal in the vibration signal of the nuclear steam turbine, hindering the ability to distinguish faults that are not easily distinguishable in the vibration signal data.
By obtaining the vibration signal of the work equipment, the signal is decomposed to obtain the target multi -modal signal, including the noise signal, and enters the preset fault classification model. Failure classification.
The clear capture of the noise signal in the vibration signal of the nuclear steam turbine has been realized, and the ability to distinguish between failure to distinguish faults is improved, and the deficiencies of physical models, signal processing and machine learning methods are avoided.
Smart Images
Figure CN119337254A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent monitoring and diagnosis technology and the cross-technical field of computer information processing, and in particular to a noise-resistant fault classification method, device, equipment, storage medium and product. Background Art
[0002] The core spirit of Industry 4.0 is to fully integrate digital technology into the economic structure, aiming to apply intelligence and automation to the entire production, management and service sectors. In this context, steam turbines, as an indispensable part of the energy conversion process, must adapt to the requirements of the new era. The combination of nuclear energy and Industry 4.0 further highlights the urgency of strengthening the fault diagnosis mechanism of nuclear steam turbines. Due to the key role of nuclear steam turbines in nuclear power generation systems, maintaining the safety and reliability of these turbines is essential to ensuring sustainable energy production and promoting environmental management.
[0003] In the related technology, nuclear steam turbine fault diagnosis methods can be roughly divided into four categories: methods based on physical models, methods relying on signal processing, methods utilizing machine learning, and hybrid methods integrating these paradigms. Physical model-based methods, signal processing-based methods, and machine learning-based methods are difficult to clearly capture the noise signals in the vibration signals of nuclear steam turbines, which hinders the ability to identify faults that are difficult to distinguish in vibration signal data. Summary of the invention
[0004] The main purpose of this application is to provide a noise-resistant fault classification method, device, equipment, storage medium and product, aiming to solve the technical problem that physical model-based methods, signal processing-based methods and machine learning-based methods are difficult to clearly capture the noise signal in the vibration signal of the nuclear steam turbine, which hinders the ability to identify faults that are difficult to distinguish in the vibration signal data.
[0005] To achieve the above objectives, the present application proposes a noise-resistant fault classification method, device, equipment, storage medium and product. The noise-resistant fault classification method includes:
[0006] Obtain vibration signals of working equipment;
[0007] Performing signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal;
[0008] The target multimodal signal and the vibration signal are input into a preset fault classification model, feature extraction is performed through a double-layer convolutional network to obtain extracted target features, and fault classification is performed through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
[0009] In one embodiment, the double-layer convolutional network includes a first convolutional network and a second convolutional network, and the step of inputting the target multimodal signal and the vibration signal into a preset fault classification model, performing feature extraction through the double-layer convolutional network to obtain extracted target features, and performing fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result includes:
[0010] Inputting the target multimodal signal and the vibration signal into the first convolutional network for feature extraction to obtain shallow features;
[0011] Inputting the target multimodal signal and the vibration signal into the second convolutional network for feature extraction to obtain deep features;
[0012] The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model for fault classification to obtain a target fault classification result.
[0013] In one embodiment, the step of performing signal decomposition on the vibration signal to obtain a target multimodal signal includes:
[0014] Decomposing the vibration signal to obtain an initial multimodal signal;
[0015] Converting the initial multimodal signal into a frequency domain signal to obtain the converted initial multimodal signal;
[0016] Performing frequency spectrum updating on the converted initial multimodal signal to obtain a target frequency spectrum in which the frequency and modal variation amplitude of the initial multimodal signal are less than a preset convergence threshold;
[0017] A weighted average frequency calculation is performed on the target frequency spectrum to obtain a target center frequency, and a target multimodal signal is determined based on the target center frequency.
[0018] In one embodiment, the step of concatenating the shallow features and the deep features, and inputting the concatenated target features into the fault classification sub-model for fault classification to obtain a target fault classification result includes:
[0019] The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model to perform feature matching to obtain an initial fault classification result;
[0020] Based on the preset target weights, weighted voting is performed on the initial fault classification results, and the category with the highest number of votes is used as the target fault classification result.
[0021] In one embodiment, the step of concatenating the shallow features and the deep features, inputting the concatenated target features into the fault classification sub-model, performing feature matching, and obtaining an initial fault classification result includes:
[0022] Performing a linear transformation on the target feature to obtain a transformed target matrix;
[0023] Based on the target matrix, the attention score of each attention head is calculated;
[0024] Based on the attention scores, determining initial outputs of different attention heads;
[0025] Concatenating the initial outputs of the different attention heads to obtain a concatenated initial output;
[0026] Based on the initial output, fault matching is performed to obtain an initial fault classification result.
[0027] In one embodiment, before the step of performing weighted voting on the initial fault classification results and taking the category with the highest number of votes as the final classification result, the following steps are included:
[0028] Obtaining a first classification result obtained by inputting a preset fault classification model into a sample modal signal and an accurate second classification result corresponding to the sample modal signal;
[0029] Obtaining initial weights, and assigning initial weights to the first classification results;
[0030] Based on the initial weights, weighted voting is performed on the first classification results to obtain a predicted third classification result;
[0031] Based on the second classification result and the third classification result, the initial weight is adjusted to obtain a preset target weight.
[0032] In addition, to achieve the above purpose, the present application also proposes a noise-resistant fault classification device, the noise-resistant fault classification device comprising:
[0033] An acquisition module, used for acquiring a vibration signal of a working device;
[0034] A decomposition module, configured to perform signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal;
[0035] The classification module is used to input the target multimodal signal and the vibration signal into a preset fault classification model, perform feature extraction through a double-layer convolutional network to obtain extracted target features, and perform fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
[0036] In addition, to achieve the above-mentioned purpose, the present application also proposes a noise-resistant fault classification device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is configured to implement the steps of the noise-resistant fault classification method as described above.
[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the noise-resistant fault classification method described above are implemented.
[0038] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the noise-resistant fault classification method as described above are implemented.
[0039] One or more technical solutions proposed in this application have at least the following technical effects:
[0040] Compared with the related technologies, the physical model-based methods, the signal processing-based methods and the machine learning-based methods are difficult to clearly capture the noise signals in the vibration signals of the nuclear steam turbines, which hinders the ability to identify the faults that are difficult to distinguish in the vibration signal data. ... The present application obtains the vibration signal of the working equipment; decomposes the vibration signal to obtain the target multimodal signal, wherein the target multimodal signal includes the noise signal; inputs the target multimodal signal and the vibration signal into a preset fault classification model, extracts features through a two-layer convolutional network to obtain the extracted target features, and classifies the faults through the fault classification sub-model of the preset fault classification model to obtain the target fault classification result. It can be understood that the present application obtains the vibration signal of the working equipment; decomposes the vibration signal to obtain the target multimodal signal, wherein the target multimodal signal includes the noise signal; inputs the target multimodal signal and the vibration signal into a preset fault classification model, extracts features through a two-layer convolutional network to obtain the extracted target features, and classifies the faults through the fault classification sub-model of the preset fault classification model to obtain the target fault classification result. The signal is decomposed into a working signal to obtain a target multimodal signal including a noise signal, and the decomposed target multimodal signal and vibration signal are input into a preset fault classification model, and a more specific feature extraction is performed through a double-layer convolutional network, and the signal is accurately classified through a fault classification sub-model, thereby avoiding the difficulty of physical model-based methods, signal processing-based methods and machine learning-based methods in clearly capturing noise signals in vibration signals of nuclear steam turbines, and then obtaining target fault classification results, thereby avoiding the difficulty of physical model-based methods, signal processing-based methods and machine learning-based methods in clearly capturing noise signals in vibration signals of nuclear steam turbines, and hindering the ability to identify faults that are difficult to distinguish in vibration signal data. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0043] Figure 1 A flowchart diagram of the first embodiment of the method for fault classification against noise provided by the present application;
[0044] Figure 2 The overall flow chart of the noise-resistant fault classification method of this application;
[0045] Figure 3 A flowchart diagram of the second embodiment of the fault classification method for noise resistance of the present application is provided;
[0046] Figure 4 This is a schematic diagram of the module structure of the noise-resistant fault classification device according to an embodiment of the present application;
[0047] Figure 5 A schematic diagram of the device structure of the hardware operating environment involved in the noise-resistant fault classification method in the embodiment of the present application.
[0048] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0049] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.
[0050] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.
[0051] The main solution of the embodiment of the present application is: obtain the vibration signal of the working equipment; perform signal decomposition on the vibration signal to obtain a target multimodal signal; input the modal signal and the noise signal into a preset fault classification model for fault classification to obtain a target fault classification result.
[0052] In the related technologies, nuclear steam turbine fault diagnosis methods can be roughly divided into four categories: methods based on physical models, methods relying on signal processing, methods utilizing machine learning, and hybrid methods that integrate these paradigms. Physical model-based methods and signal processing-based methods face challenges in understanding the complex nonlinear interactions of nuclear steam turbines, hindering the ability to identify faults that are difficult to distinguish in signal data.
[0053] The present application obtains the vibration signal of the working equipment and decomposes the working signal to obtain a target multimodal signal including a noise signal, and inputs the decomposed target multimodal signal and the vibration signal into a preset fault classification model, and classifies the signal through the architecture of a first convolutional network, a second convolutional network and a fault classification sub-model, thereby avoiding the difficulty of physical model-based methods, signal processing-based methods and machine learning-based methods in clearly capturing the noise signal in the vibration signal of the nuclear steam turbine, and then obtaining the target fault classification result. Furthermore, it can be avoided that the difficulty of physical model-based methods, signal processing-based methods and machine learning-based methods in clearly capturing the noise signal in the vibration signal of the nuclear steam turbine hinders the ability to identify faults that are difficult to distinguish in the vibration signal data.
[0054] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, etc. The following takes a fault classification device as an example to illustrate this embodiment and the following embodiments.
[0055] Based on this, the embodiment of the present application provides a noise-resistant fault classification method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the noise-resistant fault classification method of the present application.
[0056] In this embodiment, please refer to Figure 2 , Figure 2 Provided is an overall flow chart of a noise-resistant fault classification method, wherein the noise-resistant fault classification method comprises steps S100 to S300:
[0057] Step S100, obtaining a vibration signal of a working device;
[0058] It should be noted that the fault classification device is provided with a signal sensor for collecting vibration signals of the working equipment. These sensors are usually installed on key components, such as bearings and gearboxes, to monitor their status. The working equipment is a nuclear steam turbine. The fault classification device obtains the vibration signals of the nuclear steam turbine collected by the signal sensor.
[0059] Step S200, performing signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal;
[0060] It is understandable that the number of target multimodal signals is 5. The target multimodal signal has a center frequency and a bandwidth, thereby ensuring that the mode is both sparse and has a limited bandwidth in the frequency domain. The fault classification device decomposes the vibration signal into 5 different modal signals through the signal processing technology variable mode decomposition (VMD, Variational Mode Decomposition), and these modes have a center frequency and a bandwidth, thereby ensuring that the mode is both sparse and has a limited bandwidth in the frequency domain, wherein one modal signal among the 5 different modal signals is used to characterize the noise signal.
[0061] Step S300, input the target multimodal signal and the vibration signal into a preset fault classification model, perform feature extraction through a double-layer convolutional network to obtain extracted target features, and perform fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
[0062] It should be noted that the preset fault classification model is the CNN-Transformer framework. The fault classification sub-model is the Transformer model. The fault classification device inputs the target multimodal signal and vibration signal into the preset fault classification model for fault classification, performs more comprehensive feature extraction through a double-layer convolutional network, and performs accurate fault classification through the Transformer model to obtain the target fault classification result.
[0063] In particular, both the standard mode signal and the vibration signal are input into the preset fault classification model in order to prevent signal loss and perform more accurate noise feature extraction.
[0064] In a feasible implementation manner, step S300 includes the following steps:
[0065] Inputting the target multimodal signal and the vibration signal into the first convolutional network for feature extraction to obtain shallow features;
[0066] It is understandable that the first convolutional network is a shallow convolutional network (CNN). Shallow features are low-level features. The first convolutional network includes a convolutional layer, an activation function, and a pooling layer. The fault classification device extracts low-level features in the input signal through the convolutional layer of the first convolutional network.
[0067] Specifically, multiple filters (convolution kernels) of the convolution layer perform convolution operations on the input signal to extract low-level features from the input signal. The convolution operation formula is as follows:
[0068]
[0069] Among them, x is the input signal, w is the convolution kernel, and y is the output feature map.
[0070] The ReLU activation function is used to introduce nonlinear features, allowing the neural network to learn complex patterns. The ReLU activation function sets negative values to zero and keeps positive values unchanged, thereby increasing the nonlinear representation ability of the model. The activation function formula is as follows:
[0071] f(x)=max(0,x)
[0072] The pooling layer is a maximum pooling layer used to downsample the output of the convolutional layer, thereby reducing the data dimension, reducing the amount of calculation and the risk of overfitting. The pooling operation formula is as follows:
[0073] y[i,j]=max(x[i:i+f,j:j+f])
[0074] Where f is the pooling window size, usually 2 × 2. Max pooling selects the maximum value within the pooling window as the output.
[0075] Inputting the target multimodal signal and the vibration signal into the second convolutional network for feature extraction to obtain deep features;
[0076] It should be noted that the second convolutional network is a deep convolutional network (CNN). The deep features are high-level features. The second convolutional network is composed of multiple first convolutional networks. The fault classification device extracts high-level features in the input signal through the convolutional layer of the second convolutional network.
[0077] Specifically, the fault classification device repeats the operation steps of the first volume and the network to extract more advanced features.
[0078] The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model for fault classification to obtain a target fault classification result.
[0079] It is understandable that the fault classification device combines the features extracted by the shallow and deep CNNs in the fusion part by splicing and inputs them into the Transformer model to perform fault classification and obtain the target fault classification result. The formula for feature fusion is:
[0080] x=Concat(x shallow ,x deep )
[0081] Among them, x shallow represents the features extracted by the shallow CNN, x deep Represents the features extracted by deep CNN.
[0082] In a feasible implementation manner, the shallow features and the deep features are concatenated, and the concatenated target features are input into the fault classification sub-model for fault classification. The step of obtaining the target fault classification result includes the following steps:
[0083] The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model to perform feature matching to obtain an initial fault classification result;
[0084] It should be noted that the initial fault classification result is the initial fault classification result predicted by the fault classification sub-model. The fault classification device inputs the spliced target features into the fault classification sub-model, performs feature matching, and obtains the initial fault classification result predicted by the fault classification sub-model.
[0085] Based on the preset target weights, weighted voting is performed on the initial fault classification results, and the category with the highest number of votes is used as the target fault classification result.
[0086] It is understandable that the fault classification device performs weighted voting on the initial fault classification results based on the preset target weights, adjusts the relative importance of different fault categories, and takes the category with the highest weighted votes as the final target fault classification result. This means that after the weighted adjustment, the fault category with the most votes will be regarded as the most likely fault type. For example, some types of faults may be more important or more common than other types, so their weights will be higher.
[0087] In a feasible implementation manner, the shallow features and the deep features are concatenated, and the concatenated target features are input into the fault classification sub-model to perform feature matching to obtain the initial fault classification result, which includes the following steps:
[0088] Performing a linear transformation on the target feature to obtain a transformed target matrix;
[0089] It should be noted that the target matrix includes Q (Query), K (Key) and V (Value). The fault classification sub-model includes an input embedding layer, an output layer, an encoder and a decoder. The encoder is composed of multiple identical layers stacked together, and each layer usually contains a multi-head self-attention mechanism and a feedforward neural network. The fault classification device linearly transforms the features extracted by CNN in the input embedding layer so that the features can adapt to the subsequent processing steps of the Transformer, and uses different weight matrices to transform the input features for each attention head of the multi-head self-attention mechanism to obtain a target matrix including Q, K and V.
[0090] Based on the target matrix, the attention score of each attention head is calculated;
[0091] It is understandable that the attention score is a set of values calculated in the attention mechanism to measure the importance or correlation between different elements in the input sequence. The fault classification device uses a multi-head self-attention mechanism to calculate the attention score in parallel using multiple attention heads. The attention score can capture the long-distance dependencies in the input features, thereby better understanding the relationship between complex signals. The formula of the multi-head attention mechanism is as follows:
[0092]
[0093] Among them, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and d k Dimensions of the key vector.
[0094] Based on the attention scores, determining initial outputs of different attention heads;
[0095] It should be noted that the initial output is based on the attention score. After determining the attention weight, the value vector is weighted and summed to obtain the final attention output. The fault classification device converts the original attention score into a probability distribution through the softmax function, so that each element is between 0 and 1, and the sum of all elements is equal to 1. This step is to enable the model to assign different weights according to the importance of each input element, using the probability distribution obtained after softmax as the weight, and weighted summing the value vector to obtain the final attention output.
[0096] Concatenating the initial outputs of the different attention heads to obtain a concatenated initial output;
[0097] It is understandable that the fault classification device concatenates the initial outputs of different attention heads to obtain the initial output for inputting the decoder.
[0098] Based on the initial output, fault matching is performed to obtain an initial fault classification result.
[0099] It should be noted that the initial fault classification result is the result of the decoder output. After obtaining the initial output, the fault classification device uses a feedforward neural network to further process the attention output through two linear transformations and a nonlinear activation function ReLU to improve the feature representation capability. The feedforward neural network formula is as follows:
[0100] FFN(z)=max(0,zW1+b1)W2+b2
[0101] Among them, FFN represents the output of the feedforward neural network, W1 and W2 represent the weight matrices of the linear transformation, and b1 and b2 represent the bias terms of the linear transformation.
[0102] It is understandable that after the initial output passes through the feedforward neural network, the fault classification device will further process the initial output with improved feature representation capability and input it to the decoder. The decoder will linearly transform the encoder output to adapt to the decoder input requirements. Similar to the Encoder part, the multi-head attention mechanism is used to capture long-distance dependencies and the nonlinear change output of the feedforward neural network, and then the output is converted into a classification result through the softmax method.
[0103] In a feasible implementation manner, the step of performing weighted voting on the initial fault classification results and taking the category with the highest number of votes as the final classification result includes the following steps:
[0104] Obtaining a first classification result obtained by inputting a preset fault classification model into a sample modal signal and an accurate second classification result corresponding to the sample modal signal;
[0105] It should be noted that the second classification result is a known correct sample label. The fault classification device collects a set of samples, each sample contains multiple modal signals, and collects the classification results of each modal signal through a two-layer CNN-Transformer framework. The classification result is expressed as the probability distribution of each category (called the first classification result). The probability distribution formula is as follows:
[0106] P k =softmax(y k ),k=1,2,…,K
[0107] Among them, P k is the classification probability distribution of the kth modal signal, y k is the classification result of the kth modal signal.
[0108] Obtaining initial weights, and assigning initial weights to the first classification results;
[0109] It is understandable that the initial weight is a uniform weight. The fault classification device sets an initial weight, which will be used in the subsequent weighted voting process. The initial weight sets the classification result of each modal signal to have the same initial voting weight, and the set initial weight is assigned to the first classification result.
[0110] Based on the initial weights, weighted voting is performed on the first classification results to obtain a predicted third classification result;
[0111] It should be noted that the third classification result is to weight the first classification result according to the initial weight, so as to obtain a new prediction result. The fault classification device uses the initial weight to perform weighted voting on the first classification result. For each sample, its first classification result will be weighted according to the initial weight, so as to obtain a new prediction result.
[0112] Based on the second classification result and the third classification result, the initial weight is adjusted to obtain a preset target weight.
[0113] It is understandable that the target weight makes the model perform best when all modal information is comprehensively considered. The fault classification device adjusts the initial weight according to the difference between the third classification result and the true second classification result. By repeatedly iterating the above process, a set of weights is found so that when these weights are applied to the first classification result, the third classification result obtained is as consistent as possible with the second classification result, and the weights found are normalized to obtain the preset target weight. The voting system performs weighted voting according to the processing results of each signal, and adjusts the voting weights in real time to form an optimized final fault classification result. This process utilizes the comprehensive insights in the original signal and the decomposed signal to ensure the comprehensiveness and accuracy of the diagnosis results.
[0114] Specifically, the voting weight is adjusted in real time according to the classification accuracy of each modal signal. A feedback mechanism is used to evaluate the classification performance of each modal signal and adjust the weight accordingly. The formula is as follows:
[0115]
[0116] in, is the weight of the kth modal signal in the tth round, η is the learning rate, Δ k is the change in classification performance of the kth modal signal.
[0117] The adjusted weights are normalized to ensure that the sum of all weights is 1. The formula is:
[0118]
[0119] in, is the normalized weight.
[0120] In this embodiment, the fault classification device processes the modal signal through a two-layer convolutional neural network (CNN). CNN is used to extract features from each modal signal and capture important patterns and features in the signal. The two-layer CNN architecture ensures accurate feature extraction of each modal component. Through the hierarchical feature extraction process, the richness and accuracy of feature representation are improved. After CNN processing, the extracted features enter the Transformer model for further processing. The Transformer model has a powerful sequence modeling capability and can capture long-distance dependencies and complex interactions between features, thereby improving the accuracy and robustness of fault classification. In this way, the features of each modal signal are fully processed and expressed.
[0121] The present application embodiment provides a noise-resistant fault classification method, referring to Figure 3 , Figure 3 This is a flow chart of the second embodiment of the noise-resistant fault classification method of the present application.
[0122] In a feasible implementation manner, step S200 includes steps A10 to A40:
[0123] Step A10, decomposing the vibration signal to obtain an initial multimodal signal;
[0124] It should be noted that the fault classification device first initially decomposes the complex vibration signal into initial multi-modal signals.
[0125] Step A20, converting the initial multimodal signal into a frequency domain signal to obtain the converted initial multimodal signal;
[0126] It is understandable that the fault classification device performs Fourier transform on the initial multimodal signal to convert the initial multimodal signal in the time domain into the initial multimodal signal in the frequency domain.
[0127] Step A30, updating the spectrum of the converted initial multimodal signal to obtain a target spectrum in which the frequency and modal change amplitude of the initial multimodal signal are less than a preset convergence threshold;
[0128] It should be noted that the fault classification device updates the frequency spectrum of each modal signal to ensure that the representation of each modal signal in the frequency domain is more accurate.
[0129] Specifically, the fault classification device uses the Wiener filter method to update the spectrum of each mode. When the frequency and modal change amplitude of the initial multimodal signal are less than the preset convergence threshold, the update is stopped. The specific formula is as follows:
[0130]
[0131] in, is the spectrum of the kth mode at the iteration, is the positive frequency part of the signal spectrum, is the dual variable, α is the balancing parameter for data fidelity, is the center frequency of the kth mode during iteration, and the Wiener filter update updates the spectrum of each modal signal by minimizing the bandwidth.
[0132] Step A40: Perform weighted average frequency calculation on the target spectrum to obtain a target center frequency, and determine a target multimodal signal based on the target center frequency.
[0133] It is understandable that the center frequency update is used to adjust the center frequency of each modal signal so that it better represents the corresponding frequency component. The fault classification device calculates the updated center frequency by the weighted average frequency of the spectrum. The specific formula is as follows:
[0134]
[0135] Among them, the weight is Center frequency update is used to adjust the center frequency of each modal signal so that it better represents the corresponding frequency components.
[0136] In particular, after the center frequency is updated, the dual variable needs to be updated. The specific formula for updating the dual variable through the gradient ascent method is as follows:
[0137]
[0138] Where τ is the time step of gradient ascent. The dual variable update is used to ensure that the superposition of modal signals can reconstruct the original signal.
[0139] In this embodiment, the fault classification device decomposes the vibration data of the nuclear steam turbine into a plurality of different modal signals by using the variable mode decomposition (VMD) method. These modal signals respectively represent different frequency components in the vibration data, thereby effectively reducing noise and improving signal clarity.
[0140] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the anti-noise fault classification method of the present application. More simple transformations based on this technical concept are all within the protection scope of the present application.
[0141] This application also provides a noise-resistant fault classification device, please refer to Figure 4 , the noise-resistant fault classification device comprises:
[0142] An acquisition module 10 is used to acquire a vibration signal of a working device;
[0143] A decomposition module 20, configured to perform signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal;
[0144] The classification module 30 is used to input the target multimodal signal and the vibration signal into a preset fault classification model, perform feature extraction through a double-layer convolutional network to obtain extracted target features, and perform fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
[0145] Optionally, the decomposition module includes:
[0146] The weighted average module is used to decompose the vibration signal to obtain an initial multimodal signal; convert the initial multimodal signal into a frequency domain signal to obtain the converted initial multimodal signal; perform spectrum update on the converted initial multimodal signal to obtain a target spectrum in which the frequency and modal change amplitude of the initial multimodal signal are less than a preset convergence threshold; perform weighted average frequency calculation on the target spectrum to obtain a target center frequency, and determine a target multimodal signal based on the target center frequency.
[0147] Optionally, the classification module includes:
[0148] The feature extraction module is used to input the target multimodal signal and the vibration signal into the first convolutional network for feature extraction to obtain shallow features; input the target multimodal signal and the vibration signal into the second convolutional network for feature extraction to obtain deep features; splice the shallow features and the deep features, and input the spliced target features into the fault classification submodel for fault classification to obtain target fault classification results.
[0149] Optionally, the feature extraction module includes:
[0150] The matching module is used to splice the shallow features and the deep features, and input the spliced target features into the fault classification sub-model to perform feature matching to obtain an initial fault classification result; based on a preset target weight, a weighted vote is performed on the initial fault classification result, and the category with the highest number of votes is used as the target fault classification result.
[0151] Optionally, the matching module includes:
[0152] A calculation module is used to linearly transform the target features to obtain a transformed target matrix; based on the target matrix, calculate the attention score of each attention head; based on the attention score, determine the initial outputs of different attention heads; splice the initial outputs of the different attention heads to obtain a spliced initial output; based on the initial output, perform fault matching to obtain an initial fault classification result.
[0153] A weight adjustment module is used to obtain a first classification result obtained by inputting a preset fault classification model of a sample modal signal and an accurate second classification result corresponding to the sample modal signal; obtain an initial weight and assign an initial weight to the first classification result; based on the initial weight, perform a weighted vote on the first classification result to obtain a predicted third classification result; based on the second classification result and the third classification result, perform a weight adjustment on the initial weight to obtain a preset target weight.
[0154] The anti-noise fault classification device provided by the present application adopts the anti-noise fault classification method in the above embodiment, which can solve the technical problem of anti-noise fault classification. Compared with the prior art, the beneficial effects of the anti-noise fault classification device provided by the present application are the same as the beneficial effects of the anti-noise fault classification method provided by the above embodiment, and other technical features in the anti-noise fault classification device are the same as the features disclosed in the above embodiment method, which will not be repeated here.
[0155] The present application provides a noise-resistant fault classification device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the noise-resistant fault classification method in the above-mentioned embodiment one.
[0156] Reference below Figure 5 , which shows a schematic diagram of the structure of a noise-resistant fault classification device suitable for implementing the embodiment of the present application. The noise-resistant fault classification device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, tablet computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The noise-resistant fault classification device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0157] like Figure 5 As shown, the noise-resistant fault classification device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the noise-resistant fault classification device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 may allow the noise-resistant fault classification device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a noise-resistant fault classification device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.
[0158] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.
[0159] The anti-noise fault classification device provided by the present application adopts the anti-noise fault classification method in the above embodiment, which can solve the technical problem of anti-noise fault classification. Compared with the prior art, the beneficial effects of the anti-noise fault classification device provided by the present application are the same as the beneficial effects of the anti-noise fault classification method provided by the above embodiment, and the other technical features in the anti-noise fault classification device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.
[0160] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0161] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0162] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the noise-resistant fault classification method in the above-mentioned embodiment.
[0163] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.
[0164] The computer-readable storage medium may be included in the noise-resistant fault classification device; or may exist independently without being assembled into the noise-resistant fault classification device.
[0165] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the noise-resistant fault classification device, the noise-resistant fault classification device: obtains the vibration signal of the working equipment; performs signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal; inputs the target multimodal signal and the vibration signal into a preset fault classification model, performs feature extraction through a two-layer convolutional network to obtain extracted target features, and performs fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
[0166] Computer program code for performing the operations of the present application may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0167] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0168] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.
[0169] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned anti-noise fault classification method, and can solve the technical problem of anti-noise fault classification. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the anti-noise fault classification method provided by the above-mentioned embodiment, and will not be repeated here.
[0170] The present application also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the above-mentioned noise-resistant fault classification method are implemented.
[0171] The computer program product provided by the present application can solve the technical problem of noise-resistant fault classification. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the noise-resistant fault classification method provided by the above embodiment, which will not be repeated here.
[0172] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.
Claims
1. A noise-resistant fault classification method, characterized in that: The anti-noise fault classification method includes: Obtain vibration signals of working equipment; Performing signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal; The target multimodal signal and the vibration signal are input into a preset fault classification model, feature extraction is performed through a double-layer convolutional network to obtain extracted target features, and fault classification is performed through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
2. The noise-resistant fault classification method according to claim 1, characterized in that: The double-layer convolution network includes a first convolution network and a second convolution network. The step of inputting the target multimodal signal and the vibration signal into a preset fault classification model, performing feature extraction through the double-layer convolution network to obtain extracted target features, and performing fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result includes: Inputting the target multimodal signal and the vibration signal into the first convolutional network for feature extraction to obtain shallow features; Inputting the target multimodal signal and the vibration signal into the second convolutional network for feature extraction to obtain deep features; The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model for fault classification to obtain a target fault classification result.
3. The noise-resistant fault classification method according to claim 1, characterized in that: The step of performing signal decomposition on the vibration signal to obtain a target multimodal signal comprises: Decomposing the vibration signal to obtain an initial multimodal signal; Converting the initial multimodal signal into a frequency domain signal to obtain the converted initial multimodal signal; Performing frequency spectrum updating on the converted initial multimodal signal to obtain a target frequency spectrum in which the frequency and modal variation amplitude of the initial multimodal signal are less than a preset convergence threshold; A weighted average frequency calculation is performed on the target frequency spectrum to obtain a target center frequency, and a target multimodal signal is determined based on the target center frequency.
4. The noise-resistant fault classification method according to claim 2, characterized in that: The step of concatenating the shallow features and the deep features, and inputting the concatenated target features into the fault classification sub-model for fault classification to obtain the target fault classification result comprises: The shallow features and the deep features are spliced together, and the spliced target features are input into the fault classification sub-model to perform feature matching to obtain an initial fault classification result; Based on the preset target weights, weighted voting is performed on the initial fault classification results, and the category with the highest number of votes is used as the target fault classification result.
5. The noise-resistant fault classification method according to claim 4, characterized in that: The step of concatenating the shallow features and the deep features, inputting the concatenated target features into the fault classification sub-model, performing feature matching, and obtaining an initial fault classification result comprises: Performing a linear transformation on the target feature to obtain a transformed target matrix; Based on the target matrix, the attention score of each attention head is calculated; Based on the attention scores, determining initial outputs of different attention heads; Concatenating the initial outputs of the different attention heads to obtain a concatenated initial output; Based on the initial output, fault matching is performed to obtain an initial fault classification result.
6. The noise-resistant fault classification method according to claim 4, characterized in that: The step of performing weighted voting on the initial fault classification results and taking the category with the highest number of votes as the final classification result includes: Obtaining a first classification result obtained by inputting a preset fault classification model into a sample modal signal and an accurate second classification result corresponding to the sample modal signal; Obtaining an initial weight, and assigning an initial weight to the first classification result; Based on the initial weights, weighted voting is performed on the first classification results to obtain a predicted third classification result; Based on the second classification result and the third classification result, the initial weight is adjusted to obtain a preset target weight.
7. A noise-resistant fault classification device, characterized in that: The device comprises: An acquisition module, used for acquiring a vibration signal of a working device; A decomposition module, used for performing signal decomposition on the vibration signal to obtain a target multimodal signal, wherein the target multimodal signal includes a noise signal; The classification module is used to input the target multimodal signal and the vibration signal into a preset fault classification model, perform feature extraction through a double-layer convolutional network to obtain extracted target features, and perform fault classification through a fault classification sub-model of the preset fault classification model to obtain a target fault classification result.
8. A noise-resistant fault classification device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the noise-resistant fault classification method according to any one of claims 1 to 6.
9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the anti-noise fault classification method according to any one of claims 1 to 6 are implemented.
10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the steps of the noise-resistant fault classification method according to any one of claims 1 to 6.