CN114239402A - Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network - Google Patents

Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network Download PDF

Info

Publication number
CN114239402A
CN114239402A CN202111549776.5A CN202111549776A CN114239402A CN 114239402 A CN114239402 A CN 114239402A CN 202111549776 A CN202111549776 A CN 202111549776A CN 114239402 A CN114239402 A CN 114239402A
Authority
CN
China
Prior art keywords
nuclear power
water pump
capsule network
circulating water
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111549776.5A
Other languages
Chinese (zh)
Inventor
王航
邓强
张博文
彭敏俊
夏庚磊
王晨阳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202111549776.5A priority Critical patent/CN114239402A/en
Publication of CN114239402A publication Critical patent/CN114239402A/en
Priority to GB2215173.2A priority patent/GB2623358A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Hardware Design (AREA)
  • Acoustics & Sound (AREA)
  • Geometry (AREA)
  • Monitoring And Testing Of Nuclear Reactors (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention relates to a nuclear power circulating water pump fault diagnosis method and system based on an optimized capsule network, wherein the method comprises the following steps: acquiring vibration sensing data of a nuclear power circulating water pump during operation and vibration sensing data under various faults to form a first data set; preprocessing the first data set to obtain a second data set; extracting features of the second data set to obtain a feature matrix; performing phase space reconstruction on the feature matrix to obtain training data; constructing a time convolution capsule network; training the time convolution capsule network by using the training data to obtain a trained time convolution capsule network; acquiring vibration sensing data of a nuclear power circulating water pump to be detected; and judging a fault result of the nuclear power circulating water pump to be detected by utilizing the trained time convolution capsule network. The invention can improve the accuracy of fault diagnosis.

Description

Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a method and a system for diagnosing faults of a nuclear power circulating water pump based on an optimized capsule network.
Background
The nuclear power system has a complex structure, has potential radioactive release danger and has extremely high requirements on safety. Therefore, the reliability requirements for the critical equipment of the nuclear power system are very high; meanwhile, with the requirements of offshore drilling platforms, island power generation and the like, a large number of operators cannot be arranged on related platforms, so that the requirements on the automation and intelligentization level of the operation of the nuclear power plant are very high, and the requirements on unattended operation and unattended operation are strong. The nuclear power system has a severe operating environment, and the key equipment works continuously for a long time, so that faults are easy to occur, if the faults occur, the faults cannot be found and maintained in time, serious radioactive consequences can be caused, and the life safety of operating personnel and the public is critical.
The circulating water pump is indispensable equipment in the long-term operation of the nuclear power plant, can provide a driving pressure head for a working medium to take away heat energy of a reactor and accessory equipment thereof, but the fault of a bearing in the pump can be found and processed only in the regular disassembly and maintenance process, so that the safety and the economy of the nuclear power plant can be seriously threatened once the bearing breaks down in the operation, and therefore, the research on the fault diagnosis method of the bearing of the circulating water pump needs to be carried out. With the continuous development of artificial intelligence technology and big data theory, the accumulation of a large amount of operation data of the nuclear power system and application experience in other fields, the fault diagnosis is quickly and accurately carried out by adopting a plurality of efficient and accurate artificial intelligence technologies, so that the operation and maintenance guarantee capacity of key equipment of the nuclear power system can be effectively improved, operation and maintenance personnel are assisted to carry out decision analysis, and the operation safety and economy are improved.
In 1967, the mechanical failure prevention group was established by the naval research institute of the united states, from which the research work of failure diagnosis technology began, and then the research and application of failure diagnosis technology gradually spread all over the world; the development of fault diagnosis technology is further promoted by the establishment of the british association of machine health and condition monitoring at the end of the 60's of the 20 th century; subsequently, relevant researches on state monitoring and fault diagnosis technologies are also developed successively in various countries in Europe, and characteristic diagnosis technology systems are formed; the fault diagnosis technology of Japan starts to start in the middle of the 70 s, and by learning and using for reference of research and continuous improvement of various countries in the world, the fault diagnosis technology of Japan in civil industries such as steel production, railway operation, chemical process and the like is mature at present; the research related to the Chinese fault diagnosis technology starts in the early 80 s, and a relatively perfect theoretical system is formed at present. In the nuclear power field, typical research efforts include the american atton laboratory to develop operator-oriented operational decision support systems; an operating state monitoring and diagnosing system for development of the Halden reactor project of the European Union; an accident diagnosis consultation system for fault diagnosis of a nuclear power plant was developed by the korean science and technology institute; a200 MW nuclear heating station fault diagnosis system is researched and developed by Qinghua university, and a nuclear power plant operation support system is designed and developed by Harbin engineering university, wherein the system comprises the functions of state monitoring, alarm analysis, fault diagnosis, emergency operation guidance and the like.
The fault diagnosis method can be divided into three types, namely a method based on a quantitative analysis model, a method based on qualitative experience knowledge and a method based on historical data. In the aspect of a fault diagnosis method based on a quantitative analysis model, in order to solve the problem of nonlinear system faults, Wiinnenberg firstly provides a fault diagnosis method of a nonlinear unknown observer. The fault diagnosis method based on the filter is provided for a nonlinear discrete system, Julie and the like, and sigma point random distribution under the characteristic input is added, so that the fault diagnosis precision of the nonlinear filter is further improved. The equivalent space method was first proposed by Chow and Willsky in 1984; in 1997, Isermann and Balle have reviewed the analytical model-based approach in detail, as well as the equivalent space approach therein. In the 90 s, the American air force adopts an equivalent space method to realize the fault detection and separation of an aircraft control system. However, for non-linear systems, the application of such methods is severely limited due to the difficulty in building accurate mathematical models for them.
In the aspect of fault diagnosis research based on qualitative experience knowledge, the fault diagnosis method does not need to establish an analytic model of a system, and the diagnosis result is easy to understand and good in robustness; but there is a difficulty in acquisition of expert knowledge; when the rules are more, the problems of matching conflict, combination explosion and the like exist in the reasoning process. As early as 1980, expert systems were applied to fault diagnosis, which is the first time that humans have transformed past learned experience into a suite of evaluation systems for fault diagnosis. Pang et al propose a distributed-based expert system that can distribute the functionality of the expert system to multiple processors for parallel operation, thereby improving the processing efficiency of the system. BO and the like provide an object-oriented knowledge representation method aiming at the dual problems of low universality and low expandability of various available expert systems, so that fault rules of a specific machine can be solved by using general rules. Due to the fact that measuring points are limited, the acquired fault phenomenon can show ambiguity, and the introduction of the fuzzy fault method is beneficial to solving the problems of inaccurate, uncertain and noise of information and the like in detection and diagnosis. Liu et al propose to combine fuzzy measurement and fuzzy integral to analyze mechanical fault data, and have good performance in the aspect of bearing and motor fault diagnosis.
In terms of fault diagnosis based on historical data, the method has the advantage over the two methods that the data or signals can be directly processed without establishing an accurate analytical model of the core. Therefore, the method has wide universality and wide application in both linear systems and nonlinear systems. The method based on historical data mainly comprises a fault diagnosis method based on a multivariate statistical method and a signal analysis and a fault diagnosis method based on artificial intelligence and pattern recognition:
(1) multivariate-based statistical methods such as Principal Component Analysis (PCA), kernel principal component analysis, independent component analysis, and the like have been rapidly developed at the end of the last century. Misra et al propose the application of PCA and the improved method thereof in fault detection in the actual industrial process, and compared with the traditional PCA-based method, the proposed improved method MSPCA greatly reduces the false alarm rate; however, such methods are mainly applied to fault detection, and have poor effects on identifying and classifying fault causes.
(2) Fault diagnosis methods based on signal analysis started to rise in the last 80 th century, and such methods mainly include wavelet transform, hilbert-yellow transform, S transform, and the like. Wavelet transform-based methods are currently the most common and reliable methods for processing signals. Leung et al reviews the application of wavelet transforms in chemical analysis for noise cancellation and data compression in different areas of analytical chemistry. In an actual industrial process, various forms of noise generally exist in an acquired signal, and a useful signal containing the noise can be decomposed by using a method based on signal analysis to achieve the effect of distinguishing the useful signal from the noise, so that the method based on signal analysis is mainly used for data denoising, preprocessing and the like. Since the signal analysis method does not have the capability of pattern recognition and classification, the signal analysis-based method is often used in combination with the pattern recognition method.
(3) Artificial intelligence and pattern recognition based methods. As early as 1988, researchers have applied neural networks to fault diagnosis in rotating machines. The types of neural networks currently used for fault detection and diagnosis are mainly: adaptive networks, radial basis networks (RBF networks), back propagation algorithms (BP networks), etc. Venkata subramanian et al first proposed the application of BP networks to process fault diagnosis. Gome et al use a Gaussian radial basis function neural network to analyze the accident of a pressurized water reactor power plant, Sinuhe uses a strategy based on an artificial neural network to detect the reactor core assembly blockage fault of a sodium-cooled fast reactor,
Figure BDA0003417100450000031
a multi-layer neural network of the 'jump' type is provided, and two neural networks are used for dynamically identifying and verifying the identification result respectively. Besides the shallow neural network, many scholars have conducted fault diagnosis technology research by using various models such as logistic regression, support vector machine, decision tree, and the like. However, these machine learning methods need to combine artificial experience to select characteristic parameters, so that the network training stability is poor and the accuracy cannot be further improvedHigh and therefore difficult to adapt to the requirements of intelligent fault diagnosis. With the rapid development of artificial intelligence technology, the research of deep learning has been successful in the fields of image recognition, speech recognition, natural language, language translation, etc. At present, fault identification and diagnosis research based on a deep learning algorithm is still in a preliminary exploration stage on the whole. Tamiselvan et al propose a multi-sensor health diagnostic method based on a deep belief network. The Luchunshan and the like realize effective diagnosis of the faults of the refining air compressor based on the deep confidence network, and the results also show that the diagnosis accuracy and stability of the method are better than those of the traditional shallow neural network.
The invention selects artificial intelligence and pattern recognition technology from the historical data-based method to realize the intelligent fault diagnosis method. The deep learning method can avoid manually selecting the characteristic parameters, and the diagnostic result is better in stability and accuracy, so that the intelligent fault diagnosis is carried out by adopting the deep learning technology. The time convolution network is a special deep neural network, and the action principle of the time convolution network is to construct a plurality of filters to perform feature extraction on input samples through layer-by-layer convolution and pooling calculation and to mine hidden information in data layer by layer.
Each neuron in the capsule network is a vector instead of a traditional scalar, so that the capsule network can extract more detailed features from input data, and loss of feature information is reduced; the capsule network updates the parameters of the capsule layer through a dynamic routing mechanism, further increases the coupling coefficient of the child node and the father node, and fully utilizes local information to enrich the characteristic representation capability and the information content; the capsule network structure has translation invariance, the relative position relation of input features can be extracted, and the accuracy of fault diagnosis of the nuclear power equipment is improved. Thus, the capsule network is better suited to handle highly non-linear data, which the data of nuclear power cycle water pumps are conforming to.
However, when the capsule network performs dynamic routing iteration, the calculated amount is large, and in addition, the deep learning method adopts a deep structure which is several times of that of the traditional shallow machine learning model, so that the calculation efficiency is far lower than that of the shallow model, and the requirement on hardware is high. Therefore, the invention provides the support vector for optimizing the dynamic routing of the capsule network aiming at the problem, thereby effectively reducing the calculation difficulty and improving the efficiency of fault diagnosis. Finally, the time convolution capsule network provided by the invention can be used for accurately diagnosing the circulating water pump of the nuclear power device, has good stability and universality, can improve the accuracy of fault diagnosis, and finally provides analysis and reference basis for operators, thereby improving the safety and reliability of the nuclear power device.
Disclosure of Invention
The invention aims to provide a nuclear power circulating water pump fault diagnosis method and system based on an optimized capsule network, which can improve the accuracy of fault diagnosis.
In order to achieve the purpose, the invention provides the following scheme:
a nuclear power circulating water pump fault diagnosis method based on an optimized capsule network comprises the following steps:
acquiring vibration sensing data of a nuclear power circulating water pump during operation and vibration sensing data under various faults to form a first data set;
preprocessing the first data set to obtain a second data set;
extracting features of the second data set to obtain a feature matrix;
performing phase space reconstruction on the feature matrix to obtain training data;
constructing a time convolution capsule network;
training the time convolution capsule network by using the training data to obtain a trained time convolution capsule network;
acquiring vibration sensing data of a nuclear power circulating water pump to be detected;
and judging a fault result of the nuclear power circulating water pump to be detected by utilizing the trained time convolution capsule network.
Optionally, the preprocessing the first data set to obtain a second data set includes:
and carrying out noise reduction processing on the first data set through wavelet packet transformation, and simultaneously carrying out orthogonal decomposition on a low-frequency part and a high-frequency part of the first data set to obtain a second data set.
Optionally, after the step of "obtaining the vibration sensing data of the nuclear power circulating water pump during operation and the vibration sensing data under various faults to form a first data set", and before the step of preprocessing the first data set to obtain a second data set ", the method further includes:
and setting different labels for the first data set according to the fault degree.
Optionally, the constructing a time convolution capsule network includes:
constructing a layer of convolutional neural network for extracting nonlinear characteristics of the detection data;
constructing a time convolution kernel at the output end of the convolution neural network for extracting depth time sequence characteristics;
constructing a capsule network at the output end of the time convolution kernel for extracting vector characteristics;
and setting a dynamic routing algorithm in the capsule network for updating and iterating the vector characteristics.
Optionally, the time convolution capsule network uses a cross entropy loss function as the loss function.
Optionally, the time convolution capsule network is trained by using an SGD optimization algorithm.
Optionally, after the step of "obtaining vibration sensing data of the nuclear power circulating water pump to be detected", and before the step of "determining a fault result of the nuclear power circulating water pump to be detected by using the trained time convolution capsule network", the method further includes:
and preprocessing the vibration sensing data of the nuclear power circulating water pump to be detected.
Optionally, the preprocessing is performed on the vibration sensing data of the nuclear power circulating water pump to be detected, and the preprocessing includes:
and carrying out noise reduction treatment on the vibration sensing data of the nuclear power circulating water pump to be detected through wavelet packet transformation, and simultaneously carrying out orthogonal decomposition on a low-frequency part and a high-frequency part of the vibration sensing data of the nuclear power circulating water pump to be detected.
Optionally, after the step of "constructing the time convolution capsule network", and before the step of training the time convolution capsule network by using the training data to obtain the trained time convolution capsule network ", the method further includes:
and setting a hyper-parameter for the time convolution capsule network.
A nuclear power circulating water pump fault diagnosis system based on an optimized capsule network comprises:
the system comprises a first data acquisition module, a second data acquisition module and a data processing module, wherein the first data acquisition module is used for acquiring vibration sensing data when a nuclear power circulating water pump operates and vibration sensing data under various faults to form a first data set;
the preprocessing module is used for preprocessing the first data set to obtain a second data set;
the characteristic extraction module is used for extracting characteristics of the second data set to obtain a characteristic matrix;
the phase space reconstruction module is used for performing phase space reconstruction on the characteristic matrix to obtain training data;
the network construction module is used for constructing a time convolution capsule network;
the training module is used for training the time convolution capsule network by using the training data to obtain a trained time convolution capsule network;
the second data acquisition module is used for acquiring vibration sensing data of the nuclear power circulating water pump to be detected;
and the detection module is used for judging a fault result of the nuclear power circulating water pump to be detected by utilizing the trained time convolution capsule network.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention selects artificial intelligence and pattern recognition technology from the historical data-based method to realize the intelligent fault diagnosis method. The deep learning method can avoid manually selecting the characteristic parameters, and the diagnostic result is better in stability and accuracy, so that the intelligent fault diagnosis is carried out by adopting the deep learning technology. The time convolution network is a special deep neural network, and the action principle of the time convolution network is to construct a plurality of filters to perform feature extraction on input samples through layer-by-layer convolution and pooling calculation and to mine hidden information in data layer by layer.
Each neuron in the capsule network is a vector instead of a traditional scalar, so that the capsule network can extract more detailed features from input data, and loss of feature information is reduced; the capsule network updates the parameters of the capsule layer through a dynamic routing mechanism, further increases the coupling coefficient of the child node and the father node, and fully utilizes local information to enrich the characteristic representation capability and the information content; the capsule network structure has translation invariance, the relative position relation of input features can be extracted, and the accuracy of fault diagnosis of the nuclear power equipment is improved. Thus, the capsule network is better suited to handle highly non-linear data, which the data of nuclear power cycle water pumps are conforming to.
However, when the capsule network performs dynamic routing iteration, the calculated amount is large, and in addition, the deep learning method adopts a deep structure which is several times of that of the traditional shallow machine learning model, so that the calculation efficiency is far lower than that of the shallow model, and the requirement on hardware is high. Therefore, the invention provides the support vector for optimizing the dynamic routing of the capsule network aiming at the problem, thereby effectively reducing the calculation difficulty and improving the efficiency of fault diagnosis. Finally, the time convolution capsule network provided by the invention can be used for accurately diagnosing the circulating water pump of the nuclear power device, has good stability and universality, can improve the accuracy of fault diagnosis, and finally provides analysis and reference basis for operators, thereby improving the safety and reliability of the nuclear power device.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a fault diagnosis method of a nuclear power circulating water pump based on an optimized capsule network;
FIG. 2 is a block diagram of a nuclear power circulating water pump fault diagnosis system based on an optimized capsule network.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a nuclear power circulating water pump fault diagnosis method and system based on an optimized capsule network, which can improve the accuracy of fault diagnosis.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, a flow chart of a method for diagnosing a fault of a nuclear power circulating water pump based on an optimized capsule network is provided, and the method comprises the following basic steps:
step 1: the method comprises the steps of collecting and storing vibration sensing data of a nuclear power circulating water pump during operation and vibration sensing data of the nuclear power circulating water pump under various faults as a first data set.
Step 2: the collected sensing data in the first data set are managed in a classified manner in the computer according to the subsystem to which the sensor belongs, and different labels can be set for the sensing data in a normal state, a fault state and different fault degrees, so that the method is convenient for subsequent training.
And step 3: carrying out noise reduction processing on the acquired original data by wavelet packet transformation on all the data in the step 2, and simultaneously carrying out sum of low frequency parts of the dataOrthogonal decomposition is carried out on the high-frequency part to improve the time domain resolution, and then wavelet packet reconstruction is carried out, and a formula
Figure BDA0003417100450000081
And reconstructing the wavelet packet of the high-frequency coefficient pair signal obtained after the low-frequency coefficient and the threshold value are quantized after the orthogonal decomposition of the wavelet packet.
Wherein k is the transformation parameter, hl-2kIs a low-pass filter coefficient, gl-2kIn order to be a high-pass filter coefficient,
Figure BDA0003417100450000082
and
Figure BDA0003417100450000083
decomposition coefficients obtained by orthogonal decomposition for the original signal.
And 4, step 4: and (3) performing feature extraction processing on the data subjected to noise reduction in the step (3), reflecting the time series mutability of the signal components through the permutation entropy, reflecting the ordering degree of the data through the envelope entropy, adjusting and aligning according to the dimensionality of the obtained data, and combining the two to form a hybrid enhanced feature matrix.
And 5: the input data of the capsule network is at least three-dimensional data, wherein the first dimension represents the total data amount, the second dimension represents the length of single data, and the third dimension represents the width of the single data; and the data in step 4 is a two-dimensional array, wherein the first dimension represents the total amount of data, and the second dimension represents the dimension of the feature. In order to enable the data of the nuclear power circulating water pump to be input into a convolutional neural network for effective fault diagnosis, the invention carries out phase space reconstruction on the data in the step 4, wherein the interval time is set to be 1s, the length of a sliding time window is set to be 20s, and finally, two-dimensional data (N multiplied by D dimension) in the step 4 is converted into a three-dimensional stacked data block of (N-num _ steps +1) × (num _ steps multiplied by D), wherein N is the total amount of data, D is the dimension of a characteristic parameter, num _ steps is the length of the sliding time window, and as the data are overlapped in each sliding process, the total data input length is (N-num _ steps +1) for the algorithm disclosed by the invention.
Step 6: the method comprises the steps of firstly establishing a layer of convolutional neural network, and preliminarily extracting the nonlinear characteristics of the measured data. The method is characterized by comprising an input layer (training data after pretreatment) and a pooling layer.
The convolution layer adopts the formula of equation (1) to extract features, after convolution operation, the feature graph needs to be fed forward and output to the pooling layer through an activation function, wherein k is a convolution kernel, b is a bias parameter, and x isjIs the output of the convolutional layer, yjAs output of the convolutional layer, MjAnd forming a characteristic map for the data obtained in the step 5.
The invention adopts the Leaky ReLU activation function, can avoid dead nodes on the basis of the ReLU activation function, and can embody the nonlinear characteristics in the data; the pooling layer is calculated using equation (2), where xjAs output of the pooling layer, yjFor the input of the pooling layer, down is a pooling function, beta is the network multiplicative bias of the l layer of the pooling layer, and b is the bias; the invention adopts maximum pooling calculation, and pooling operation can perform down-sampling on training data to prevent model over-fitting.
Figure BDA0003417100450000091
xj＝βjdown(yj)+bj (2)
And 7: extracting depth time sequence characteristics by adopting a time convolution kernel, increasing the receptive field by expansion convolution, and inputting the sequence X { X ] obtained in the step 61,x2,…,xnIn which x1～xnIs the quantity in sequence X, the dilation convolution is:
Figure BDA0003417100450000092
where d represents the dilation factor, k represents the filter size, s-d i represents the past direction, f (i) represents the convolution kernel, and f(s) is the output of the dilation convolution. And use residual convolution to avoid the gradient vanishing problem:
o＝Activation(X+F(X)) (4)
where X is the result of the dilated convolution in equation (3) as the input to the killing convolution, o is the output, f (X) represents the output of the last hidden layer through the residual network, and Activation is the Activation function. The present invention employs a two-layer time convolution structure.
And 8: the characteristic information is input into the capsule network to extract vector characteristics, and loss of key characteristic information is reduced. Multiplying the output characteristics of the time convolution layer by the weight matrix to obtain a prediction vector:
Uj,i＝UiWj,i (5)
wherein Wj,iIs the weight of the main capsule layer, UiIs a characteristic of the time convolution layer output, Uj,iRepresenting the vector generated by the input feature prediction. Transmitting the prediction vector to the digital capsule layer:
Figure BDA0003417100450000093
Sj＝∑iUj,iCi,j (7)
Figure BDA0003417100450000094
bi,j＝bi,j+VjUj,i (9)
in the formula Ci,jAnd bi,jRepresenting the coupling and bias coefficients, SjIs the total input vector.
And step 9: aiming at the problem of large iterative calculation amount of the dynamic routing from the formula (6) to the formula (9) in the capsule network in the step 8, the invention provides a dynamic routing algorithm optimized by a support vector for the first time. Will Uj,iAfter being input into a support vector machine for training, a group of support vectors sv can be obtained1,sv2，…，svQAnd with Lagrange factor a1,a2，…，aNLagrange vector a of elements1×NIso-feature information and parameters. Reconstructing the original training sample by a feature extraction formula shown in formula (10):
Figure BDA0003417100450000101
in the formula sviIs a support vector, aiIt is the support vector that corresponds to the lagrangian factor,
Figure BDA0003417100450000102
is the support vector corresponds tag, b is the offset.
The reconstructed sample number is kept consistent with the initial training sample set to be N, and each sample dimension is changed from the initial sample dimension M to Q. For training sample xi∈Uj,iEquation (10) for input xiAfter processing, a new set of reconstructed samples can be obtained as shown in equation (11):
Figure BDA0003417100450000103
then h obtained by reconstructioniAs a new Uj,iAnd substituting the dynamic route for updating iteration.
Step 10: combing hyper-parameters in a time-convolutional capsule network; 5, 6, 7 and 8, relating to a large number of hyper-parameters in the process of setting the structure of the time convolution capsule network, wherein the hyper-parameters comprise the number of middle hidden layers of the convolution neural network, the convolution kernel size of convolution layers, the step length of the convolution process and the number of characteristic graphs; the size of the convolution kernel of the time convolution kernel, the step size and the sparse rate of the convolution process; the input capsule number, the input vector dimension, the output capsule number, the output vector dimension and the routing iteration number of the capsule network; penalty coefficients, activation functions, etc. of the support vector machine.
Step 11: defining a loss function and optimizing parameters; the invention adopts a cross entropy loss function as a loss function. In order to optimize the weight and the bias in the time convolution capsule network, an SGD optimization algorithm is adopted to solve the network in the training process so as to minimize the value of a loss function.
Meanwhile, in the training process of the time convolution capsule network, all data are split into a plurality of batches of training samples, each batch has 32 groups of data, and processed data are randomly disturbed to reduce uncertainty and prevent overfitting. With the increase of the number of training rounds, the training error is gradually reduced, which shows that the time convolution capsule network model can continuously approach the parameter variation characteristic under the actual fault.
Step 12: and in the actual fault diagnosis process, preprocessing the abnormal data according to the steps 1-5 to ensure that the data processing mode is completely consistent with the training data.
Step 13: and (4) diagnosing typical faults of the circulating water pump of the nuclear power plant by using the time convolution capsule network model optimized in the step (11) to obtain a classification result.
In addition, the method can also evaluate the fault diagnosis result of the model, and the accuracy and the effectiveness of the model are evaluated by using the confusion matrix and the fault diagnosis accuracy as indexes. The related results can be referred by operation and decision-making personnel, and related measures can be taken in time, so that the safety is ensured, and the economy can be improved.
It should be noted that the fault diagnosis of the nuclear power plant circulating water pump can be performed through machine learning such as a support vector machine and a back propagation neural network and a common convolutional neural network technology, but the method provided by the invention can effectively extract the time sequence characteristic and the depth vector characteristic of input data by combining a time convolutional network and a capsule network algorithm, and can obtain a more accurate fault diagnosis result aiming at the characteristic of high linearity of the nuclear power plant circulating water pump operation data. The operating data of the circulating water pump is subjected to noise reduction through wavelet packet transformation, so that the influence of noise on the data is reduced; performing feature extraction on the denoised data through the permutation entropy and the envelope entropy to obtain a mixed enhanced feature; extracting high-dimensional time sequence characteristics of the obtained data through time convolution kernel mining characteristics; the relative position relation between the vector characteristics and the data is mined through the capsule network, and the dynamic routing of the capsule network is optimized through the support vector, so that the efficiency of the capsule network is improved. Finally, the method can adaptively, accurately and quickly diagnose the potential fault cause in the circulating water pump of the nuclear power plant, and provides analysis and reference basis for operators.
Based on the method, the invention also provides a nuclear power circulating water pump fault diagnosis system based on the optimized capsule network, as shown in fig. 2, comprising:
the first data acquisition module 201 is used for acquiring vibration sensing data of the nuclear power circulating water pump during operation and vibration sensing data under various faults to form a first data set;
a preprocessing module 202, configured to preprocess the first data set to obtain a second data set;
a feature extraction module 203, configured to perform feature extraction on the second data set to obtain a feature matrix;
a phase space reconstruction module 204, configured to perform phase space reconstruction on the feature matrix to obtain training data;
a network construction module 205 for constructing a time convolution capsule network;
a training module 206, configured to train the time convolution capsule network with the training data to obtain a trained time convolution capsule network;
the second data acquisition module 207 is used for acquiring vibration sensing data of the nuclear power circulating water pump to be detected;
and the detection module 208 is configured to determine a fault result of the nuclear power circulating water pump to be detected by using the trained time convolution capsule network.
Based on the above, the invention also discloses the following technical effects:
compared with the prior art, the method has higher fault diagnosis accuracy than other methods, can obtain better convergence effect under the condition of smaller training data set, and can provide a fault diagnosis model with higher accuracy and better convergence effect.
The reason for the high accuracy is the overall implementation of step 3, step 4, step 5, step 7, step 8, step 9, and step 10.
And 3, denoising the original data through wavelet packet transformation in the steps 3 and 4, reducing the influence of environmental noise on the data when the circulating water pump operates, and performing feature extraction processing on the denoised data through the permutation entropy and the envelope entropy to form a hybrid enhanced feature, so as to provide rich and accurate data support for a subsequent fault diagnosis model.
In the step 5, original 2-dimensional data is converted into a three-dimensional data group with time sequence attributes, so that the input data format of the time convolution capsule network can be satisfied as far as possible, and each input data is not only a single instantaneous parameter but also a period of time sequence characteristics, and the data characteristic change of the fault process can be reflected better.
Step 7, establishing a time convolution network structure formed by stacking time convolution kernels, and compared with the traditional convolution kernels, the time sequence information characteristics of the data can be more effectively extracted; meanwhile, the flexible receptive field can be flexibly customized according to different characteristics of different tasks; and a more stable gradient is provided, and the problems of gradient disappearance and explosion are avoided.
Each neuron in the capsule network formed in the step 8 is a vector instead of a traditional scalar, so that the capsule network can extract more detailed features from input data, and the loss of feature information is reduced; the capsule network updates the parameters of the capsule layer through a dynamic routing mechanism, further increases the coupling coefficient of the child node and the father node, and fully utilizes local information to enrich the characteristic representation capability and the information content; the capsule network structure has translation invariance, and can extract the relative position relation of input features, thereby improving the accuracy of fault diagnosis.
And 9, aiming at the problems that the capsule network dynamic routing calculation amount is large and the diagnosis efficiency is not high enough, the provided support vector dynamic routing algorithm can self-adaptively acquire key feature points in input data through a support vector machine, reconstruct the data, enhance the feature resolution of the data, accelerate the convergence of a capsule network model and improve the efficiency and accuracy of fault diagnosis.
Step 10, comprehensively combing the hyper-parameters needing to be given manually in the time convolution capsule network, and selecting the optimal parameters aiming at model test adjustment to ensure the fault diagnosis accuracy and the network stability.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. A nuclear power circulating water pump fault diagnosis method based on an optimized capsule network is characterized by comprising the following steps:
acquiring vibration sensing data of a nuclear power circulating water pump during operation and vibration sensing data under various faults to form a first data set;
preprocessing the first data set to obtain a second data set;
extracting features of the second data set to obtain a feature matrix;
performing phase space reconstruction on the feature matrix to obtain training data;
constructing a time convolution capsule network;
training the time convolution capsule network by using the training data to obtain a trained time convolution capsule network;
acquiring vibration sensing data of a nuclear power circulating water pump to be detected;
and judging a fault result of the nuclear power circulating water pump to be detected by utilizing the trained time convolution capsule network.
2. The method for diagnosing the fault of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 1, wherein the preprocessing the first data set to obtain a second data set comprises:
and carrying out noise reduction processing on the first data set through wavelet packet transformation, and simultaneously carrying out orthogonal decomposition on a low-frequency part and a high-frequency part of the first data set to obtain a second data set.
3. The method for diagnosing the faults of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 1, wherein after the step of obtaining the vibration sensing data of the nuclear power circulating water pump during operation and the vibration sensing data under various faults to form a first data set, and before the step of preprocessing the first data set to obtain a second data set, the method further comprises the following steps of:
and setting different labels for the first data set according to the fault degree.
4. The method for diagnosing the fault of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 1, wherein the step of constructing the time convolution capsule network comprises the following steps:
constructing a layer of convolutional neural network for extracting nonlinear characteristics of the detection data;
constructing a time convolution kernel at the output end of the convolution neural network for extracting depth time sequence characteristics;
constructing a capsule network at the output end of the time convolution kernel for extracting vector characteristics;
and setting a dynamic routing algorithm in the capsule network for updating and iterating the vector characteristics.
5. The optimized capsule network-based nuclear power circulating water pump fault diagnosis method according to claim 1, wherein the time convolution capsule network adopts a cross entropy loss function as the loss function.
6. The optimized capsule network-based nuclear power circulating water pump fault diagnosis method according to claim 1, wherein the time convolution capsule network is trained by adopting an SGD optimization algorithm.
7. The method for diagnosing the faults of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 1, wherein after the step of obtaining the vibration sensing data of the nuclear power circulating water pump to be detected and before the step of judging the fault result of the nuclear power circulating water pump to be detected by using the trained time convolution capsule network, the method further comprises the following steps of:
and preprocessing the vibration sensing data of the nuclear power circulating water pump to be detected.
8. The method for diagnosing the fault of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 7, wherein the preprocessing of the vibration sensing data of the nuclear power circulating water pump to be detected comprises:
and carrying out noise reduction treatment on the vibration sensing data of the nuclear power circulating water pump to be detected through wavelet packet transformation, and simultaneously carrying out orthogonal decomposition on a low-frequency part and a high-frequency part of the vibration sensing data of the nuclear power circulating water pump to be detected.
9. The method for diagnosing the faults of the nuclear power circulating water pump based on the optimized capsule network as claimed in claim 1, wherein after the step of constructing the time convolution capsule network, and before the step of training the time convolution capsule network by using the training data to obtain the trained time convolution capsule network, the method further comprises:
and setting a hyper-parameter for the time convolution capsule network.
10. A nuclear power circulating water pump fault diagnosis system based on an optimized capsule network is characterized by comprising the following components:
the system comprises a first data acquisition module, a second data acquisition module and a data processing module, wherein the first data acquisition module is used for acquiring vibration sensing data when a nuclear power circulating water pump operates and vibration sensing data under various faults to form a first data set;
the preprocessing module is used for preprocessing the first data set to obtain a second data set;
the characteristic extraction module is used for extracting characteristics of the second data set to obtain a characteristic matrix;
the phase space reconstruction module is used for performing phase space reconstruction on the characteristic matrix to obtain training data;
the network construction module is used for constructing a time convolution capsule network;
the training module is used for training the time convolution capsule network by using the training data to obtain a trained time convolution capsule network;
the second data acquisition module is used for acquiring vibration sensing data of the nuclear power circulating water pump to be detected;
and the detection module is used for judging a fault result of the nuclear power circulating water pump to be detected by utilizing the trained time convolution capsule network.
CN202111549776.5A 2021-12-17 2021-12-17 Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network Pending CN114239402A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202111549776.5A CN114239402A (en) 2021-12-17 2021-12-17 Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network
GB2215173.2A GB2623358A (en) 2021-12-17 2022-10-14 Method and system for fault diagnosis of nuclear power circulating water pump based on optimized capsule network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111549776.5A CN114239402A (en) 2021-12-17 2021-12-17 Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network

Publications (1)

Publication Number Publication Date
CN114239402A true CN114239402A (en) 2022-03-25

Family

ID=80758103

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111549776.5A Pending CN114239402A (en) 2021-12-17 2021-12-17 Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network

Country Status (2)

Country Link
CN (1) CN114239402A (en)
GB (1) GB2623358A (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115929613A (en) * 2022-12-21 2023-04-07 中国联合网络通信集团有限公司 Method, device, system and medium for processing faults of oil transfer pump

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255882B (en) * 2021-04-30 2022-11-04 南通大学 Bearing fault diagnosis method based on improved convolution capsule network
CN113669246B (en) * 2021-08-23 2022-06-24 水利部交通运输部国家能源局南京水利科学研究院 Intelligent diagnosis method for water pump fault under cross-working condition

Also Published As

Publication number Publication date
GB202215173D0 (en) 2022-11-30
GB2623358A (en) 2024-04-17

Similar Documents

Publication Publication Date Title
Shao et al. A novel deep autoencoder feature learning method for rotating machinery fault diagnosis
CN110779746B (en) Diagnosis method for improving composite fault of deep sparse self-encoder network rotating machinery
Wang et al. A method for rapidly evaluating reliability and predicting remaining useful life using two-dimensional convolutional neural network with signal conversion
Cheng et al. Autoencoder quasi-recurrent neural networks for remaining useful life prediction of engineering systems
Sun et al. Fast bearing fault diagnosis of rolling element using Lévy Moth-Flame optimization algorithm and Naive Bayes
Wang et al. A Bayesian inference-based approach for performance prognostics towards uncertainty quantification and its applications on the marine diesel engine
CN111899905B (en) Fault diagnosis method and system based on nuclear power device
CN104616033A (en) Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine)
CN108519768A (en) A kind of method for diagnosing faults analyzed based on deep learning and signal
CN112016251B (en) Nuclear power device fault diagnosis method and system
Zhang et al. Knowledge distilling based model compression and feature learning in fault diagnosis
CN117290800B (en) Timing sequence anomaly detection method and system based on hypergraph attention network
Qian et al. A comparative study of deep learning-based fault diagnosis methods for rotating machines in nuclear power plants
CN111881627A (en) Nuclear power device fault diagnosis method and system
CN115791174B (en) Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium
Zhang et al. Intelligent machine fault diagnosis using convolutional neural networks and transfer learning
CN114091504A (en) Rotary machine small sample fault diagnosis method based on generation countermeasure network
CN113705396A (en) Motor fault diagnosis method, system and equipment
Shajihan et al. CNN based data anomaly detection using multi-channel imagery for structural health monitoring
CN116842379A (en) Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models
Ma et al. A collaborative central domain adaptation approach with multi-order graph embedding for bearing fault diagnosis under few-shot samples
Xu et al. Global attention mechanism based deep learning for remaining useful life prediction of aero-engine
CN114239402A (en) Nuclear power circulating water pump fault diagnosis method and system based on optimized capsule network
Zhao et al. A capsnet-based fault diagnosis method for a digital twin of a wind turbine gearbox
CN117113170A (en) Lightweight rotary machine fault diagnosis method based on multi-scale information fusion

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination