CN111428788A - Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor - Google Patents

Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor Download PDF

Info

Publication number
CN111428788A
CN111428788A CN202010214917.7A CN202010214917A CN111428788A CN 111428788 A CN111428788 A CN 111428788A CN 202010214917 A CN202010214917 A CN 202010214917A CN 111428788 A CN111428788 A CN 111428788A
Authority
CN
China
Prior art keywords
sequence
model
rotor
fault
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.)
Granted
Application number
CN202010214917.7A
Other languages
Chinese (zh)
Other versions
CN111428788B (en
Inventor
谢永慧
孙磊
张荻
郑召利
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.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong 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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202010214917.7A priority Critical patent/CN111428788B/en
Publication of CN111428788A publication Critical patent/CN111428788A/en
Application granted granted Critical
Publication of CN111428788B publication Critical patent/CN111428788B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Control Of Turbines (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a steam turbine generator set rotor multi-fault diagnosis method and system based on deep learning, which comprises the following steps: acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding, adding labels, and constructing and acquiring training set data and verification set data; constructing a sequence-to-sequence model based on a recurrent neural network; the sequence is put into a sequence model, a set decoder is used as an agent for reinforcement learning, and after a complete label sequence is generated, a reward is obtained; the training target of the sequence-to-sequence model is set as a minimum negative expected value, and the improved sequence-to-sequence model is obtained; and when the training sequence reaches the preset requirement of the sequence model, obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set.

Description

Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor
Technical Field
The invention belongs to the technical field of mechanical fault diagnosis, and relates to a multi-fault diagnosis method for a steam turbine generator set rotor, in particular to a multi-fault diagnosis method and a multi-fault diagnosis system for the steam turbine generator set rotor based on deep learning.
Background
The rotor of the steam turbine generator set is an important part for power production, has the characteristics of complex structure, high temperature, high pressure and high rotating speed, and is easy to break down; once the fault occurs, the fault cannot be checked in time, so that the fault can lead to the unplanned shutdown to cause economic loss, and can lead to the damage of a unit and the casualties at heavy time. Therefore, the fault diagnosis of the rotor of the steam turbine generator set has very important significance for guaranteeing the safe operation of the steam turbine generator set.
Steam turbine rotor faults include rotor cracks, unbalance, misalignment, rubbing, oil film instability, rotor loosening, and compound faults compounded by the above faults, and have complications. The traditional fault diagnosis method requires that a user has rich prior knowledge, and the diagnosis accuracy is low; the artificial intelligence algorithm is realized without the need of rich priori knowledge of users, and fault characteristics can be directly mined from data, so that fault classification and quantitative identification are carried out.
The problems of the conventional fault diagnosis method include:
1. the conventional fault diagnosis is only single-class identification, and the diagnosis of multi-class composite faults needs further improvement according to the situation; if the multi-class recognition is simply converted into a plurality of single-class recognition tasks, the training cost is undoubtedly increased, and the diagnosis effect is poor;
2. in the multi-class composite faults, the faults generally have higher correlation, for example, oil film instability is more easily compounded with rotor looseness, and the conventional fault diagnosis has no correlation among the fault classes, so that the waste of calculation resources is caused, and even the set effect cannot be achieved;
3. in the multi-class composite fault diagnosis process, the output fault class is essentially a non-ordered set rather than an ordered sequence, and if only the label of the label is changed into a multi-fault label for identification by a traditional method, the workload of data labeling is increased, the data amount is unequal, the diagnosis effect is affected, and therefore the sensitivity of the label sequence needs to be reduced in the diagnosis process.
In summary, a new method for diagnosing multiple faults of a steam turbine generator set rotor based on deep learning is needed.
Disclosure of Invention
The invention aims to provide a steam turbine generator set rotor multi-fault diagnosis method and system based on deep learning, and aims to solve one or more technical problems. The core problem to be solved by the invention is that the existing fault diagnosis technology is difficult to identify multiple faults and does not consider the correlation among the faults and the sensitivity of a label sequence, so that the defects of consumption of computing resources and the like exist; the invention can obtain an accurate fault identification model by depending on the correlation among faults, realize the rapid identification of multiple types of faults, and simultaneously, the identification result is irrelevant to the label sequence, thereby reducing the occupation of computing resources and the workload of data marking.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention discloses a steam turbine generator set rotor multi-fault diagnosis method based on deep learning, which comprises the following steps of:
step 1, acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding through a multi-point acquisition mode, adding labels, and constructing and acquiring training set data and verification set data; wherein the typical steam turbine generator set rotor fault comprises: rotor cracking, unbalance, misalignment, rubbing, oil film instability and rotor loosening;
step 2, constructing a sequence-to-sequence model based on a recurrent neural network; the sequence-to-sequence model includes an encoder and an ensemble decoder; each training set data fault signal is connected to a Softmax activation function through an encoder and a set decoder to obtain distribution probability; obtaining ordered labels according to the distribution probability; wherein, an attention mechanism is added in the process of obtaining the distribution probability;
step 3, the sequence is put into a sequence model, a set decoder is used as a reinforcement learning agent, and after a complete label sequence is generated, a reward r is obtained;
the training target for the sequence-to-sequence model is set to minimize negative expectation, expressed as,
Figure BDA0002424072920000021
wherein the reward r is F generating a label y and a background label y1The score, expressed as: r (y) ═ F1(y,y*)；
Obtaining an improved sequence-to-sequence model;
step 4, training the sequence-to-sequence model obtained in the step 3 by obtaining training set data in the step 1, verifying the sequence-to-sequence model obtained by training by using the verification set data obtained in the step 1, and obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set when the preset requirement is met, wherein in the training process of the sequence-to-sequence model, a variable learning rate is set to train a neural network, and the optimization L (theta) is optimized, the optimization algorithm adopts a self-judging strategy gradient algorithm, the gradient calculation expression is as follows,
Figure BDA0002424072920000031
in the formula, ysSequence of labels sampled for the probability distribution p, ygA sequence of tags generated for a greedy search algorithm.
The invention further improves the method and also comprises the following steps:
step 5, when the fault types are added, setting a small learning rate for the main structure of the cyclic neural network in the model to learn on the basis of the diagnosis model obtained in the step 4, and retraining the final full-connection layer;
and (4) when new measuring point data is added, taking the diagnosis model obtained in the step (4) as a pre-training model, and then training the network.
The invention is further improved in that the step 1 specifically comprises the following steps:
step 1.1, distributing and controlling multiple vibration measuring points, and acquiring vibration signals X of a steam turbine generator set rotor by using a displacement sensor under the condition of the same sampling ratei(ii) a Wherein, the data X of each vibration measuring pointiObtaining the data under the condition of uniform sampling rate at the same moment; x represents a long time sequence signal, i represents the ith vibration measuring point;
step 1.2, each long time sequence signal XiDivided into m short-time signals x of length lm,iThe label of the corresponding fault category is ym,iIf the fault is a composite fault, the fault is a multi-label vector; x of different vibration measuring points i under the same mm,iHigh dimensional signal (x) reconstructed as i rows and l columnsi,l)mThe label of the corresponding fault category is ym,i(ii) a Wherein x isi,lThe time sequence signal data of i rows and l columns are represented, x represents a short time sequence signal after segmentation, y represents a fault type, l represents the number of data points of a vibration signal, and m represents an m-th sample signal;
step 1.3, processing the data obtained in the step 1.2 by using a mean standard deviation formula; dividing according to a preset proportion and scrambling data to obtain training set data { (x)i,l)m}train、{ym,i}trainAnd verification set data { (x)i,l)m}validation、 {ym,i}validation。
The invention has the further improvement that the step 2 specifically comprises the following steps:
let the input be x ═ x1,…,xT) The output is (y ═ y)1,…,yT)；
Step 2.1, the encoder receives the input vector and the state value of the hidden layer at the last moment, ht＝RNNenc(xt,ht-1)，xtRepresents the input vector, htRepresenting the state value, RNN, of the hidden layer at time t in the encoderencAn RNN module adopted by the encoder is shown;
step 2.2, the set decoder receives the output vector and the state value, s, of the hidden layer at the previous momentt＝RNNdec(yt-1,st-1)， yt-1Representing the output vector, stState value, RNN, representing hidden layers at time t in an aggregate decoderdecAn RNN module employed by the set decoder is represented;
step 2.3, carrying out weighted average on the hidden layer in the encoder; firstly, calculating a score according to the hidden layer value of the set decoder and the hidden layer value of the encoder: e.g. of the typei,j＝score(si,hj) (ii) a Then computing hidden layer value weight according to the fraction:
Figure BDA0002424072920000041
and finally, performing weighted average on the hidden layer in the Encoder according to the weight:
Figure BDA0002424072920000042
step 2.4, processing the state values in the set decoder according to the weighted average values obtained in step 2.3:
Figure BDA0002424072920000043
Wcthe weight to be trained;
step 2.5, calculating the final output probability according to the softmax function:
Figure BDA0002424072920000044
wherein, in the calculation of hidden layer value of aggregate decoder and hidden layer value fraction e of encoderi,jThe attention mechanism is added.
The further improvement of the invention is that in step 4, when the turbine generator set rotor to be tested has multiple faults, the method comprises the following steps:
preprocessing and standardizing the vibration data of the turbine generator set rotor to be detected, which is acquired by a power plant, by using the mode of the step 1 to obtain processed data;
and inputting the processed data into a trained diagnosis model, outputting a fault category sequence by the model, and obtaining the multi-fault type to which the rotor of the turbine generator set to be tested belongs.
The invention discloses a steam turbine generator set rotor multi-fault diagnosis system based on deep learning, which comprises the following components:
the data acquisition module is used for acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding through a multi-point acquisition mode, adding labels, and constructing and acquiring training set data and verification set data; wherein the typical steam turbine generator set rotor fault comprises: rotor cracking, unbalance, misalignment, rubbing, oil film instability and rotor loosening;
the model building module is used for building a sequence-to-sequence model based on the recurrent neural network; the sequence-to-sequence model includes an encoder and an ensemble decoder; each training set data fault signal is connected to a Softmax activation function through an encoder and a set decoder to obtain distribution probability; obtaining ordered labels according to the distribution probability; wherein, an attention mechanism is added in the process of obtaining the distribution probability;
the model improvement module is used for taking the set decoder as a reinforcement learning agent in the sequence-to-sequence model to generate a complete label sequence and then obtain a reward r;
the training target for the sequence-to-sequence model is set to minimize negative expectation, expressed as,
wherein the reward r is F generating a label y and a background label y1The score, expressed as: r (y) ═ F1(y,y*)；
Obtaining an improved sequence-to-sequence model;
the fault diagnosis module is used for obtaining a sequence-to-sequence model obtained by the training set data training model improvement module through the data obtaining module, verifying the sequence-to-sequence model obtained by training through the verification set data obtained by the data obtaining module, and obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set when the preset requirement is met, wherein in the training process of the sequence-to-sequence model, a variable learning rate is set to train a neural network and optimize L (theta), the optimization algorithm adopts a self-judging strategy gradient algorithm, and the gradient calculation expression is,
Figure BDA0002424072920000051
in the formula, ysSequence of labels sampled for the probability distribution p, ygA sequence of tags generated for a greedy search algorithm.
Compared with the prior art, the invention has the following beneficial effects:
1. most of the existing fault diagnosis technologies are single-type fault identification, the label sequence is directly used as a model to be output, so that multi-fault identification can be realized, redundant operation is not needed in data processing, and the popularization is good;
2. the correlation between faults is not considered in the current fault identification, and the correlation between fault label sequences can be fully considered by adopting a sequence-to-sequence model based on RNN, so that the calculation cost is greatly reduced;
3. the existing fault identification model is sensitive to the sequence among the label sequences, and a reinforced learning mechanism is introduced into the Decoder part in the model, so that the influence of the label sequences on the prediction result can be reduced, and the actual production needs are met.
In conclusion, the deep learning-based multi-fault diagnosis method for the steam turbine generator set rotor can effectively diagnose various fault types, has the characteristics of small model size, high speed and strong mobility, is slightly influenced by data labels, and is easy to popularize.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art are briefly introduced below; it is obvious that the drawings in the following description are some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
Fig. 1 is a schematic flow chart of a steam turbine generator set rotor multi-fault diagnosis method based on deep learning according to an embodiment of the invention.
Detailed Description
In order to make the purpose, technical effect and technical solution of the embodiments of the present invention clearer, the following clearly and completely describes the technical solution of the embodiments of the present invention with reference to the drawings in the embodiments of the present invention; it is to be understood that the described embodiments are only some of the embodiments of the present invention. Other embodiments, which can be derived by one of ordinary skill in the art from the disclosed embodiments without inventive faculty, are intended to be within the scope of the invention.
Referring to fig. 1, a steam turbine generator set rotor multi-fault diagnosis method based on deep learning according to an embodiment of the present invention is a steam turbine rotor fault diagnosis method based on L STM, and specifically includes the following steps:
step 1, collecting fault signals at multiple measuring points and preprocessing the fault signals, wherein the method comprises the following steps: the method comprises the steps of obtaining typical steam turbine generator set rotor faults including rotor cracks, unbalance, misalignment, rubbing, oil film instability, rotor looseness and composite fault signals formed by compounding the faults, and integrating data and adding labels.
Preferably, step 1 specifically comprises:
step 1.1, distributing and controlling multiple measuring points, and acquiring vibration signals X of a steam turbine rotor by using a displacement sensor under the condition of the same sampling ratei(ii) a Wherein, the data X of each vibration measuring pointiObtaining the data under the condition of uniform sampling rate at the same moment; x represents a long time sequence signal, and i represents the ith vibration measuring point.
Step 1.2, every vibration signal X with long time sequenceiDivided into m short-time signals x of length lm,iThe label of the corresponding fault category is ym,iIf the fault is a composite fault, the fault is a multi-label vector; finally, x of different measuring points i under the same mm,iHigh dimensional signal (x) reconstructed i rows and l columnsi,l)mThe label of the corresponding fault category is ym,i(ii) a Wherein xi,lTiming signal data indicating i rows and l columns; where x represents the short timing signal after segmentation, y represents the fault category, l represents the number of data points of the vibration signal, and m represents the mth sample signal.
And 1.3, carrying out standardization processing on the data, and processing the data by using a mean-standard deviation formula. Dividing the training set/test set according to the proportion of 5-10, and disordering the training set data to obtain the training set data { (x)i,l)m}train、 {ym,i}trainAnd verification set data { (x)i,l)m}validation、{ym,i}validation。
Step 2, establishing a sequence-to-sequence model based on RNN; wherein the sequence-to-sequence model consists of an encoder E and a set decoder D, each training data signal { (x)i,l)m}trainThe distribution probability is obtained by connecting Encoder and Decode to Softmax activation function, wherein an attention mechanism is added.
Preferably, step 2 specifically comprises:
let the input be x ═ x1,…,xT) The output is (y ═ y)1,…,yT). (for { (x) in step 1i,l)m}train、{ym,i}trainIdentification is simplified).
a) Encoder accepts input vector and state value h of hidden layer at last momentt＝RNNenc(xt,ht-1)，xtRepresents the input vector, htRepresenting the state value, RNN, of the hidden layer at time t in EncoderencAn RNN module adopted by an Encoder is shown;
b) the Decoder accepts the output vector and the state value, s, of the hidden layer at the last momentt＝RNNdec(yt-1,st-1)，yt-1Representing the input vector, stState value, RNN, representing hidden layer at time t in DecoderdecAn RNN module adopted by an Encoder is shown;
c) weighted average is carried out on the hidden layer in the Encode, firstly, the hidden layer value of the Decode is calculatedHidden layer value calculation score of Encoder: e.g. of the typei,j＝score(si,hj) Then, a hidden-value weight is calculated from the score:
Figure BDA0002424072920000081
then, carrying out weighted average on the hidden layer in the Encoder according to the weight:
Figure BDA0002424072920000082
d) and processing the state value in the Decoder according to the weighted average value:
Figure BDA0002424072920000083
wherein WcThe weight to be trained;
e) the final output probability is calculated according to the softmax function:
Figure BDA0002424072920000084
wherein, the hidden layer value of Decoder and the hidden layer value fraction e of Encode are calculatedi,jAn attention mechanism is added.
And 3, obtaining the correlation of the labels according to the sequence-to-sequence model, wherein the model also depends on the sequence of the labels. To reduce dependency, the Decoder is used as a proxy for reinforcement learning, and t time can generate a tag sequence (y)1,…,yt-1). The prediction of the next label is finished by a random strategy of the Decoder, and the reward r can be obtained after a complete label sequence is generated.
The training objective was to minimize negative expectation:
Figure BDA0002424072920000086
the optimization algorithm adopts a self-criticizing strategy gradient algorithm, wherein y issSequence of labels sampled for the probability distribution p, ygTag sequences generated for a greedy search algorithm:
Figure BDA0002424072920000085
the reward r here is F of the generated label and the background label1And (3) fractional: r (y) ═ F1(y,y*)。
Step 4, training the network: training the network and implementing multiple fault diagnosis applications.
Setting variable learning rate to train the neural network according to the strategy, optimizing L (theta) and obtaining a usable diagnosis model;
and (3) inputting the vibration data of the steam turbine rotor acquired by the power plant by using a diagnosis model, preprocessing and standardizing by using the mode of the step (1), and outputting a fault category sequence by using the model, namely the multi-fault type to which the steam turbine rotor belongs.
Step 5, model maintenance:
when a new category needs to be added, setting a small learning rate for the main structure of the network on the basis of the existing model to learn, and then retraining the final full-connection layer;
when new measurement data needs to be added, the network is retrained by using the existing model as a pre-training model.
The invention provides a steam turbine generator set rotor multi-fault diagnosis system based on deep learning, which comprises the following steps:
the data acquisition module is used for acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding through a multi-point acquisition mode, adding labels, and constructing and acquiring training set data and verification set data; wherein the typical steam turbine generator set rotor fault comprises: rotor cracking, unbalance, misalignment, rubbing, oil film instability and rotor loosening;
the model building module is used for building a sequence-to-sequence model based on the recurrent neural network; the sequence-to-sequence model includes an encoder and an ensemble decoder; each training set data fault signal is connected to a Softmax activation function through an encoder and a set decoder to obtain distribution probability; obtaining ordered labels according to the distribution probability; wherein, an attention mechanism is added in the process of obtaining the distribution probability;
the model improvement module is used for taking the set decoder as a reinforcement learning agent in the sequence-to-sequence model to generate a complete label sequence and then obtain a reward r;
the training target for the sequence-to-sequence model is set to minimize negative expectation, expressed as,
Figure BDA0002424072920000091
wherein the reward r is F generating a label y and a background label y1The score, expressed as: r (y) ═ F1(y,y*)；
Obtaining an improved sequence-to-sequence model;
the fault diagnosis module is used for obtaining a sequence-to-sequence model obtained by the training set data training model improvement module through the data obtaining module, verifying the sequence-to-sequence model obtained by training through the verification set data obtained by the data obtaining module, and obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set when the preset requirement is met, wherein in the training process of the sequence-to-sequence model, a variable learning rate is set to train a neural network and optimize L (theta), the optimization algorithm adopts a self-judging strategy gradient algorithm, and the gradient calculation expression is,
Figure BDA0002424072920000101
in the formula, ysSequence of labels sampled for the probability distribution p, ygA sequence of tags generated for a greedy search algorithm.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can make modifications and equivalents to the embodiments of the present invention without departing from the spirit and scope of the present invention, which is set forth in the claims of the present application.

Claims (6)

1. A steam turbine generator set rotor multi-fault diagnosis method based on deep learning is characterized by comprising the following steps:
step 1, acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding through a multi-point acquisition mode, adding labels, and constructing and acquiring training set data and verification set data; wherein the typical steam turbine generator set rotor fault comprises: rotor cracking, unbalance, misalignment, rubbing, oil film instability and rotor loosening;
step 2, constructing a sequence-to-sequence model based on a recurrent neural network; the sequence-to-sequence model includes an encoder and an ensemble decoder; each training set data fault signal is connected to a Softmax activation function through an encoder and a set decoder to obtain distribution probability; obtaining ordered labels according to the distribution probability; wherein, an attention mechanism is added in the process of obtaining the distribution probability;
step 3, the sequence is put into a sequence model, a set decoder is used as a reinforcement learning agent, and after a complete label sequence is generated, a reward r is obtained;
the training target for the sequence-to-sequence model is set to minimize negative expectation, expressed as,
Figure FDA0002424072910000011
wherein the reward r is F generating a label y and a background label y1The score, expressed as: r (y) ═ F1(y,y*)；
Obtaining an improved sequence-to-sequence model;
step 4, training the sequence-to-sequence model obtained in the step 3 by obtaining training set data in the step 1, verifying the sequence-to-sequence model obtained by training by using the verification set data obtained in the step 1, and obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set when the preset requirement is met, wherein in the training process of the sequence-to-sequence model, a variable learning rate is set to train a neural network, and the optimization L (theta) is optimized, the optimization algorithm adopts a self-judging strategy gradient algorithm, the gradient calculation expression is as follows,
Figure FDA0002424072910000012
in the formula, ysSequence of labels sampled for the probability distribution p, ygA sequence of tags generated for a greedy search algorithm.
2. The deep learning based multiple fault diagnosis method for the steam turbine generator set rotor according to claim 1, characterized by further comprising the following steps:
step 5, when the fault types are added, setting a small learning rate for the main structure of the cyclic neural network in the model to learn on the basis of the diagnosis model obtained in the step 4, and retraining the final full-connection layer;
and (4) when new measuring point data is added, taking the diagnosis model obtained in the step (4) as a pre-training model, and then training the network.
3. The deep learning-based multi-fault diagnosis method for the steam turbine generator set rotor according to claim 1, wherein the step 1 specifically comprises the following steps:
step 1.1, distributing and controlling multiple vibration measuring points, and acquiring vibration signals X of a steam turbine generator set rotor by using a displacement sensor under the condition of the same sampling ratei(ii) a Wherein, the data X of each vibration measuring pointiObtaining the data under the condition of uniform sampling rate at the same moment; x represents a long time sequence signal, i represents the ith vibration measuring point;
step 1.2, each long time sequence signal XiDivided into m short-time signals x of length lm,iThe label of the corresponding fault category is ym,iIf the fault is a composite fault, the fault is a multi-label vector; x of different vibration measuring points i under the same mm,iHigh dimensional signal (x) reconstructed as i rows and l columnsi,l)mCorresponding to the factBarrier class label ym,i(ii) a Wherein x isi,lThe time sequence signal data of i rows and l columns are represented, x represents a short time sequence signal after segmentation, y represents a fault type, l represents the number of data points of a vibration signal, and m represents an m-th sample signal;
step 1.3, processing the data obtained in the step 1.2 by using a mean standard deviation formula; dividing according to a preset proportion and scrambling data to obtain training set data { (x)i,l)m}train、{ym,i}trainAnd verification set data { (x)i,l)m}validation、{ym,i}validation。
4. The deep learning-based multi-fault diagnosis method for the steam turbine generator set rotor according to claim 3, wherein the step 2 specifically comprises the following steps:
let the input be x ═ x1,…,xT) The output is (y ═ y)1,…,yT)；
Step 2.1, the encoder receives the input vector and the state value of the hidden layer at the last moment, ht＝RNNenc(xt,ht-1)，xtRepresents the input vector, htRepresenting the state value, RNN, of the hidden layer at time t in the encoderencAn RNN module adopted by the encoder is shown;
step 2.2, the set decoder receives the output vector and the state value, s, of the hidden layer at the previous momentt＝RNNdec(yt-1,st-1)，yt-1Representing the output vector, stState value, RNN, representing hidden layers at time t in an aggregate decoderdecAn RNN module employed by the set decoder is represented;
step 2.3, carrying out weighted average on the hidden layer in the encoder; firstly, calculating a score according to the hidden layer value of the set decoder and the hidden layer value of the encoder: e.g. of the typei,j＝score(si,hj) (ii) a Then computing hidden layer value weight according to the fraction:
Figure FDA0002424072910000031
and finally, performing weighted average on the hidden layer in the Encoder according to the weight:
Figure FDA0002424072910000032
step 2.4, processing the state values in the set decoder according to the weighted average values obtained in step 2.3:
Figure FDA0002424072910000033
Wcthe weight to be trained;
step 2.5, calculating the final output probability according to the softmax function:
Figure FDA0002424072910000034
wherein, in the calculation of hidden layer value of aggregate decoder and hidden layer value fraction e of encoderi,jThe attention mechanism is added.
5. The method for diagnosing the multiple faults of the rotor of the steam turbine generator set based on the deep learning of the claim 3 is characterized in that in the step 4, the method for diagnosing the multiple faults of the rotor of the steam turbine generator set to be tested comprises the following steps:
preprocessing and standardizing the vibration data of the turbine generator set rotor to be detected, which is acquired by a power plant, by using the mode of the step 1 to obtain processed data;
and inputting the processed data into a trained diagnosis model, outputting a fault category sequence by the model, and obtaining the multi-fault type to which the rotor of the turbine generator set to be tested belongs.
6. The utility model provides a many fault diagnosis system of steam turbine generating set rotor based on deep learning which characterized in that includes:
the data acquisition module is used for acquiring single fault signals of rotor faults of a typical steam turbine generator set and composite fault signals formed by compounding through a multi-point acquisition mode, adding labels, and constructing and acquiring training set data and verification set data; wherein the typical steam turbine generator set rotor fault comprises: rotor cracking, unbalance, misalignment, rubbing, oil film instability and rotor loosening;
the model building module is used for building a sequence-to-sequence model based on the recurrent neural network; the sequence-to-sequence model includes an encoder and an ensemble decoder; each training set data fault signal is connected to a Softmax activation function through an encoder and a set decoder to obtain distribution probability; obtaining ordered labels according to the distribution probability; wherein, an attention mechanism is added in the process of obtaining the distribution probability;
the model improvement module is used for taking the set decoder as a reinforcement learning agent in the sequence-to-sequence model to generate a complete label sequence and then obtain a reward r;
the training target for the sequence-to-sequence model is set to minimize negative expectation, expressed as,
Figure FDA0002424072910000041
wherein the reward r is F generating a label y and a background label y1The score, expressed as: r (y) ═ F1(y,y*)；
Obtaining an improved sequence-to-sequence model;
the fault diagnosis module is used for obtaining a sequence-to-sequence model obtained by the training set data training model improvement module through the data obtaining module, verifying the sequence-to-sequence model obtained by training through the verification set data obtained by the data obtaining module, and obtaining a trained diagnosis model for diagnosing multiple faults of the rotor of the steam turbine generator set when the preset requirement is met, wherein in the training process of the sequence-to-sequence model, a variable learning rate is set to train a neural network and optimize L (theta), the optimization algorithm adopts a self-judging strategy gradient algorithm, and the gradient calculation expression is,
Figure FDA0002424072910000042
in the formula, ysSequence of labels sampled for the probability distribution p, ygA sequence of tags generated for a greedy search algorithm.
CN202010214917.7A 2020-03-24 2020-03-24 Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor Active CN111428788B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010214917.7A CN111428788B (en) 2020-03-24 2020-03-24 Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010214917.7A CN111428788B (en) 2020-03-24 2020-03-24 Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor

Publications (2)

Publication Number Publication Date
CN111428788A true CN111428788A (en) 2020-07-17
CN111428788B CN111428788B (en) 2022-12-09

Family

ID=71549344

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010214917.7A Active CN111428788B (en) 2020-03-24 2020-03-24 Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor

Country Status (1)

Country Link
CN (1) CN111428788B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733887A (en) * 2020-12-24 2021-04-30 浙江大学 Method for detecting fault of hub motor of electric vehicle driven by supervision data
CN113566953A (en) * 2021-09-23 2021-10-29 中国空气动力研究与发展中心设备设计与测试技术研究所 Online monitoring method for flexible-wall spray pipe
CN113588266A (en) * 2021-07-19 2021-11-02 西安交通大学 Rolling bearing composite fault diagnosis method with embedded fault semantic space

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108197648A (en) * 2017-12-28 2018-06-22 华中科技大学 A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models
CN108960077A (en) * 2018-06-12 2018-12-07 南京航空航天大学 A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network
CN110046409A (en) * 2019-03-29 2019-07-23 西安交通大学 A kind of components of steam turbine health state evaluation method based on ResNet
CN110334764A (en) * 2019-07-04 2019-10-15 西安电子科技大学 Rotating machinery intelligent failure diagnosis method based on integrated depth self-encoding encoder

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108197648A (en) * 2017-12-28 2018-06-22 华中科技大学 A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models
CN108960077A (en) * 2018-06-12 2018-12-07 南京航空航天大学 A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network
CN110046409A (en) * 2019-03-29 2019-07-23 西安交通大学 A kind of components of steam turbine health state evaluation method based on ResNet
CN110334764A (en) * 2019-07-04 2019-10-15 西安电子科技大学 Rotating machinery intelligent failure diagnosis method based on integrated depth self-encoding encoder

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
张冉等: "改进BP网络在汽轮发电机组故障诊断中的应用", 《计算机仿真》 *
梁平等: "基于ARMA及神经网络的汽轮机振动故障诊断研究", 《热能动力工程》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733887A (en) * 2020-12-24 2021-04-30 浙江大学 Method for detecting fault of hub motor of electric vehicle driven by supervision data
CN113588266A (en) * 2021-07-19 2021-11-02 西安交通大学 Rolling bearing composite fault diagnosis method with embedded fault semantic space
CN113566953A (en) * 2021-09-23 2021-10-29 中国空气动力研究与发展中心设备设计与测试技术研究所 Online monitoring method for flexible-wall spray pipe
CN113566953B (en) * 2021-09-23 2021-11-30 中国空气动力研究与发展中心设备设计与测试技术研究所 Online monitoring method for flexible-wall spray pipe

Also Published As

Publication number Publication date
CN111428788B (en) 2022-12-09

Similar Documents

Publication Publication Date Title
CN109580215B (en) Wind power transmission system fault diagnosis method based on deep generation countermeasure network
Shao et al. Highly accurate machine fault diagnosis using deep transfer learning
Shao et al. Generative adversarial networks for data augmentation in machine fault diagnosis
CN111428788B (en) Deep learning-based multi-fault diagnosis method and system for steam turbine generator set rotor
CN105738109B (en) Bearing fault classification diagnosis method based on rarefaction representation and integrated study
Dai et al. Machinery health monitoring based on unsupervised feature learning via generative adversarial networks
Cheng et al. Autoencoder quasi-recurrent neural networks for remaining useful life prediction of engineering systems
CN112067294A (en) Rolling bearing intelligent fault diagnosis method based on deep learning
Fu et al. Adaptive broad learning system for high-efficiency fault diagnosis of rotating machinery
CN114544155A (en) AUV propeller multi-information-source fusion fault diagnosis method and system based on deep learning
CN112819037B (en) Fault diagnosis method based on classification parameter distribution of cross attention and self attention
CN115859077A (en) Multi-feature fusion motor small sample fault diagnosis method under variable working conditions
CN112393934B (en) Wind turbine generator fault diagnosis method based on sparse self-coding and extreme learning machine
CN116593157A (en) Complex working condition gear fault diagnosis method based on matching element learning under small sample
Du et al. DCGAN based data generation for process monitoring
Zhao et al. A neural architecture search method based on gradient descent for remaining useful life estimation
Li et al. Intelligent fault diagnosis of aeroengine sensors using improved pattern gradient spectrum entropy
CN115587290A (en) Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network
Bao et al. A transformer model-based approach to bearing fault diagnosis
CN114239397A (en) Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning
Techane et al. Rotating machinery prognostics and application of machine learning algorithms: Use of deep learning with similarity index measure for health status prediction
CN114021469B (en) Method for monitoring one-stage furnace process based on mixed sequence network
CN114818811A (en) Aircraft engine rolling bearing fault diagnosis method based on twin network metric learning
CN113887666A (en) Defect detection method and system for composite material based on Transformer
Zhang et al. Application of improved parallel LSTM in bearing fault diagnosis

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
GR01 Patent grant
GR01 Patent grant