CN114548154A - Intelligent diagnosis method and device for important service water pump - Google Patents

Intelligent diagnosis method and device for important service water pump Download PDF

Info

Publication number
CN114548154A
CN114548154A CN202210072680.2A CN202210072680A CN114548154A CN 114548154 A CN114548154 A CN 114548154A CN 202210072680 A CN202210072680 A CN 202210072680A CN 114548154 A CN114548154 A CN 114548154A
Authority
CN
China
Prior art keywords
fault
service water
noise reduction
encoder
sparse
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
CN202210072680.2A
Other languages
Chinese (zh)
Inventor
张荣勇
智一凡
李奇
张文杰
代丽
李娜
黄倩
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.)
China Nuclear Power Engineering Co Ltd
Original Assignee
China Nuclear Power Engineering Co Ltd
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 China Nuclear Power Engineering Co Ltd filed Critical China Nuclear Power Engineering Co Ltd
Priority to CN202210072680.2A priority Critical patent/CN114548154A/en
Publication of CN114548154A publication Critical patent/CN114548154A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E30/00—Energy generation of nuclear origin

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (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)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Signal Processing (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention relates to a method and a device for intelligently diagnosing an important service water pump, wherein the method comprises the following steps: s1: collecting vibration signals of important service water pumps under different working conditions to obtain sample points containing the vibration signals in different fault states; s2: carrying out wavelet threshold denoising pretreatment on the sample points; s3: establishing a fault game model to provide an interactive environment for observing, acting and obtaining rewards for a fault diagnosis agent; s4: dimensionality reduction and feature extraction are carried out on a plurality of hidden layers through a stacked self-coding neural network, initial parameters are optimized through a BP neural network, and a deep neural network model with the functions of feature extraction and pattern recognition is obtained; s5: and outputting the diagnosis result. The invention can effectively establish a deep neural network model by combining reinforcement learning and deep learning according to vibration signals in different states and by designing a fault diagnosis game environment, realizes intelligent diagnosis and has better diagnosis effect.

Description

Intelligent diagnosis method and device for important service water pump
Technical Field
The invention relates to the field of mechanical fault diagnosis and computer artificial intelligence, in particular to a method and a device for intelligently diagnosing an important service water pump.
Background
The important service water pump is important nuclear safety third-level equipment of a nuclear power station, and the pump is equipment serving as a heat exchanger of a cooling water system of equipment of the nuclear power station. The seawater delivered by the important plant water pump transfers the heat in the heat exchanger to the nature (sea) to ensure the safe and reliable operation of various devices of the nuclear power plant. And the performance of the water pump is reduced after the water pump fails, and huge economic loss and even disastrous accidents can be caused.
In the prior art, a common fault diagnosis method for a rotary machine performs preprocessing on data by acquiring signal data and performing empirical mode decomposition, local mean decomposition, wavelet packet decomposition and other methods. After the raw data is preprocessed, feature extraction and fault identification are required. The deep learning method can automatically learn representative characteristics from data and is successfully applied to the field of mechanical fault diagnosis. However, the deep learning model still has the disadvantages: (1) a direct linear or non-linear mapping between the raw data and the corresponding failure modes cannot be established and the performance of these failure diagnosis methods depends on the quality of building and training the deep model. (2) The training mechanism of this approach is mainly based on supervised learning or semi-supervised learning, which means that the diagnostic algorithm requires an expert system in order to learn the different failure modes exclusively.
By designing a fault diagnosis game environment and combining reinforcement learning and deep learning, a human agent can directly learn knowledge and experience from data, and the method can successfully use DRL (deep reinforcement learning) to obtain an intelligent fault diagnosis agent. The method can effectively establish an unsupervised diagnosis model, realize intelligent diagnosis and have better diagnosis effect.
Disclosure of Invention
The invention aims to provide an important intelligent diagnosis method and device for a service water pump, which have the advantages of simplicity, high efficiency, stability and accuracy, aiming at the defects of the prior art.
The technical scheme of the invention is as follows: an intelligent diagnosis method for important service water pumps comprises the following steps:
(S1) collecting vibration signals of the important service water pumps under different working conditions to obtain sample points containing the vibration signals of different fault states;
(S2) performing wavelet threshold denoising pre-processing on the sample points;
(S3) building a fault game model to provide an interactive environment for fault diagnosis agents to observe, act on and obtain rewards;
(S4) reducing dimensions and extracting features of a plurality of hidden layers by stacking self-coding neural networks, and optimizing initial parameters by using a BP neural network to obtain a deep neural network model with the functions of feature extraction and pattern recognition;
(S5) outputting the diagnosis result.
Further, in the method for intelligently diagnosing the essential service water pump as described above, in the step (S1), the vibration signals in different fault states are collected through the measurement system composed of the acceleration sensor and the LabVIEW software.
Further, in the step (S1), the cause of the failure includes: the impeller is characterized in that severe abrasion and cavitation occur at the position of the orifice ring, the center of the rotor is not coincident with the center of the volute, the blades are not properly aligned with the diffuser, the design of an inlet and outlet pipeline is unreasonable, the inlet and outlet pipeline has no straight pipe section or the straight pipe section is too short (particularly, the inlet influence is large), so that the turbulence before the fluid enters the impeller is caused, the blades do not operate under the designed working condition (such as operate under ultra-small flow or ultra-large flow), and the fluid is easy to generate turbulence, defluidization, impact, random uneven impact of cavitation on the blades and the like.
Further, in the method for intelligently diagnosing the important service water pump as described above, the method for performing denoising preprocessing on the sample point in the step (S2) includes the following steps:
(S201) denoising the noise data by improving a threshold function, and selecting an improved threshold th as follows:
Figure BDA0003482687960000031
wherein σ2And the mean is the median of the absolute value of the high-frequency wavelet coefficient, n is the signal length, j is the decomposition layer number of the wavelet, and beta is a control factor.
(S202) extracting wavelet packet energy of de-noising data by utilizing wavelet packet transformation;
(S203) carrying out wavelet packet decomposition on the vibration signal in the fault state, wherein the selected wavelet basis is db3, and the decomposition layer number is 4;
(S204) the wavelet packet energy of each node after decomposition is obtained according to the following formula:
Ejn＝1M∑t＝1M((Xjn(t))2
wherein, Xjn(t) represents a time-domain vibration signal, t represents time, and M is a node XjnThe number of samples in (c).
Further, in the method for intelligently diagnosing the important service water pump as described above, the specific process of the step (S3) includes:
(S301) generating a fault diagnosis problem;
(S302) accepting the diagnostic question via the agent and giving a result feedback;
(S303) judging whether the diagnosis result is correct for the game model, if the correct total return is added with 1, outputting a game return value, otherwise, subtracting 1 from the total return, and returning to (S301) to regenerate the fault diagnosis problem.
Further, in the method for intelligently diagnosing the important service water pump as described above, the specific process of the step (S4) includes:
(S401) adding damage noise to the preprocessed data, and adding noise reduction limitation and sparsity limitation to an automatic encoder to obtain a sparse noise reduction self-encoder;
(S402) pre-training the sparse noise reduction self-encoder;
(S403) taking the hidden layer output value of the previous sparse noise reduction self-encoder as the input value of the next sparse noise reduction self-encoder, and repeating the step (S402) by adopting a layer-by-layer greedy algorithm until all the sparse noise reduction self-encoders are trained;
(S404) adding a softmax classifier to the sparse noise reduction self-encoder for classification;
(S405) updating the weight and the bias in each iteration by using a BP back propagation algorithm, and finely adjusting the whole depth network parameter;
(S406) testing the algorithm classification accuracy by using the test set;
(S407) completing the deep neural network model.
Further, in the method for intelligently diagnosing the important service water pump, in the step (S401), the noise reduction limitation adopts mask noise as a correction, that is, the input data is randomly set to 0, and the corrected input vector is recorded as x', and the expression is as follows:
h＝f(x′)＝Sf(Wx′+b)
in the formula: x' is the modified input vector; sfRepresenting a non-linear activation function; w is an encoder weight matrix; b is the encoder bias.
Sparsity limitation is achieved by controlling the average activation amount of hidden layer neurons. Suppose aj(x) Representing the jth activation unit of the hidden layer, the average activation amount of the jth unit of the hidden layer is:
Figure BDA0003482687960000041
in the formula: n denotes the total number of samples, and j denotes the hidden layer unit.
To ensure pjWithout deviating from p, one needs to be added to the cost functionSparse term penalty factor
Figure BDA0003482687960000042
In the formula: s2Representing the number of hidden layer neurons; rho is a sparsity parameter; KL is Kullback-Leibler (induced sparsity) divergence; rhojRepresenting the average activation of the jth cell of the hidden layer.
After the noise reduction limitation is performed on the stacked self-encoder (AE) and the sparse penalty factor is added, a sparse noise reduction self-encoder (SDAE) is obtained, and then the cost function primitive can be expressed as:
JsDAE＝JAE+γPN
wherein, gamma represents the weight of the sparse penalty factor; j. the design is a squareAEIs a stacked autoencoder cost function; PN represents a sparse term penalty factor.
Further, in the method for intelligently diagnosing the important service water pump, in the step (S402), the sparse noise reduction self-encoder updates the connection weight and the bias according to the following formula in the pre-training process:
Figure BDA0003482687960000051
Figure BDA0003482687960000052
in the formula: ε represents the learning rate, JSDAERepresenting a sparse noise reduction autoencoder cost function, Wij(l) Representing the weight between the ith and jth neurons of the l-th layer, bi(l) Representing the offset between the ith layer and the ith neuron,
Figure BDA0003482687960000053
the difference is indicated.
Further, in the method for intelligently diagnosing the important service water pump as described above, the specific process of the step (S5) includes: and the display unit in the computer equipment gives an early warning on the fault type corresponding to the vibration signal and stores the fault type in the storage unit in time.
The invention further provides a device for realizing the intelligent diagnosis method of the important service water pump, which comprises the following steps:
the acceleration sensor is used for acquiring vibration signals of different fault types;
the data acquisition unit is used for isolating and normalizing input voltage and converting analog input quantity into digital quantity required by a microcomputer system so as to interface with a CPU (central processing unit);
the signal processing unit is used for carrying out wavelet threshold denoising pretreatment on the acquired signals;
the intelligent diagnosis unit is used for establishing a fault game model and a deep neural network model, intelligently identifying fault signals and automatically judging fault types by adopting the deep neural network;
the display unit is used for labeling the classified fault types on a human-computer interaction interface, and warning the corresponding fault types in time after the fault occurs;
the storage unit is used for instantly storing data and reading and writing data;
and the reset unit is used for starting the current CPU and the running data after being cleared under the condition of no power failure.
The invention has the following beneficial effects:
1. the method has the advantages that the improved threshold function is adopted to denoise important service water pump fault samples, denoised wavelet packet energy is extracted, fault information is retained to the maximum extent, and accurate diagnosis of fault data can be realized after sparse denoising self-encoder (SDAE) is carried out.
2. By designing a fault diagnosis game environment, the method can successfully use the DRL to obtain the intelligent fault diagnosis agent, can effectively establish a diagnosis model, greatly reduces the dependence on prior knowledge and expert experience of fault diagnosis on the premise of ensuring the diagnosis precision and reliability, and has better diagnosis effect.
3. The invention integrates the sparse noise reduction self-coding deep neural network to be applied to intelligent fault diagnosis, and remarkably improves the classification accuracy. The sparse noise reduction self-coding network excavates the characteristics of original data in an unsupervised learning mode, and the robustness and the generalization of the model are improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced 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 based on these drawings without creative efforts.
FIG. 1 is a flow chart of a method for intelligently diagnosing an important service water pump according to the present invention;
FIG. 2 is a flow chart of a wavelet improved threshold denoising method provided by the present invention;
FIG. 3 is a flow chart of a method for establishing a failure game model according to the present invention;
FIG. 4 is a flow chart of a method for building a deep neural network model according to the present invention;
fig. 5 is a schematic view showing a specific deconstruction of an important intelligent diagnosis device for a service water pump according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention considers that the types of the faults of the important service water pumps are more, and the traditional supervision mode is not only time-consuming and labor-consuming and has low diagnosis precision, so that a fault game model mode is designed, reinforcement learning and deep learning are combined, fault signals are preprocessed, a plurality of hidden layers are subjected to dimension reduction and feature extraction through a stacked self-coding neural network, initial parameters are optimized by using a BP neural network, and a deep neural network model with the functions of feature extraction and pattern recognition is obtained. The method can effectively establish an unsupervised diagnosis model, realize intelligent diagnosis and has better diagnosis effect.
The method for intelligently diagnosing the important service water pump, provided by the embodiment of the invention, as shown in fig. 1, comprises the following steps:
s1) acquiring vibration signals of important service water pumps under different working conditions to obtain sample points containing the vibration signals in different fault states;
s2) preprocessing the sample points by using a wavelet improved threshold denoising method;
s3), establishing a fault game model to provide an interactive environment for observing, acting and obtaining rewards for a fault diagnosis agent;
s4) dimension reduction and feature extraction are carried out on a plurality of hidden layers through a stacked self-coding neural network, initial parameters are optimized through a BP neural network, and a deep neural network model with the functions of feature extraction and pattern recognition is obtained;
s5) outputs the diagnosis result.
In step S1, the vibration signals in different fault states are obtained by a measurement system composed of an acceleration sensor and a LabVIEW; causes of the failure include: the impeller is characterized in that severe abrasion and cavitation occur at the position of the orifice ring, the center of the rotor is not coincident with the center of the volute, the blades are not properly aligned with the diffuser, the design of an inlet and outlet pipeline is unreasonable, the inlet and outlet pipeline has no straight pipe section or the straight pipe section is too short (particularly, the inlet influence is large), so that the turbulence before the fluid enters the impeller is caused, the blades do not operate under the designed working condition (such as operate under ultra-small flow or ultra-large flow), and the fluid is easy to generate turbulence, defluidization, impact, random uneven impact of cavitation on the blades and the like.
In an alternative embodiment, the preprocessing may employ a modified wavelet thresholding method; as shown in fig. 2, the method comprises the following steps:
s201) denoising the noise data by improving a threshold function;
further, in order to solve the problem that the hard threshold function is discontinuous and the soft threshold function has certain deviation, the invention provides a new threshold function between the soft threshold and the hard threshold, and the threshold function combines the advantages of the soft threshold and the hard threshold.
Aiming at the defects of the traditional fixed threshold, the invention provides an improved threshold:
Figure BDA0003482687960000081
wherein σ2And the mean is the median of the absolute value of the high-frequency wavelet coefficient, n is the signal length, j is the decomposition layer number of the wavelet, and beta is a control factor. The threshold value is gradually decreased as the number of decomposition layers increases.
S202) extracting wavelet packet energy of de-noising data by utilizing wavelet packet transformation;
s203) carrying out wavelet packet decomposition on the fault signal, wherein the selected wavelet basis is db3, and the number of decomposition layers is 4;
s204), the wavelet packet energy of each node after decomposition can be obtained by the following formula:
Ejn＝1M∑t＝1M((Xjn(t))2
wherein, Xjn(t) represents a time-domain vibration signal, t represents time, and M is XjnThe number of samples in (c). After decomposition, the number of nodes sigma iN2 is obtainediAnd i represents the number of the neurons, and N is the decomposition layer number, so that the energy of the corresponding node is obtained.
In an alternative embodiment, a failure game model may be established; as shown in fig. 3, the method comprises the following steps:
s301) generating a fault diagnosis problem;
s302) receiving the diagnosis problem through the agent and giving result feedback;
s303) judging whether the diagnosis result is correct for the game model, and if the diagnosis result is correct, adding 1 to the total return, outputting a game return value; otherwise, the total return is reduced by 1, and the step S301) is returned to regenerate the fault diagnosis problem.
In an alternative embodiment, obtaining the deep neural network model, as shown in fig. 4, includes the following steps:
s401) adding damage noise to the data after denoising pretreatment;
noise reduction and sparsity restrictions are added to the autoencoder, which can improve the robustness and generalization of the diagnostic method.
In practice, the noise reduction limit usually adopts mask noise as a correction, i.e. randomly setting 0 to the input data. Noting the corrected input vector as x', the equation is:
h＝f(x′)＝Sf(Wx′+b)
in the formula: x' is the modified input vector; sfRepresenting a non-linear activation function; w is an encoder weight matrix; b is the encoder bias.
Sparsity limitation is achieved by controlling the average activation amount of hidden layer neurons. Suppose aj(x) Representing the jth activation unit of the hidden layer, the average activation amount of the jth unit of the hidden layer is:
Figure BDA0003482687960000091
in the formula: n denotes the total number of samples, and j denotes the hidden layer unit.
To ensure pjNo deviation of rho, a sparse term penalty factor needs to be added into the cost function
Figure BDA0003482687960000092
In the formula: s2Representing the number of hidden layer neurons; rho is a sparsity parameter; KL is Kullback-Leibler (induced sparsity) divergence; rhojRepresenting the average activation of the jth cell of the hidden layer.
After the noise reduction limitation is performed on the stacked self-encoder (AE) and the sparse penalty factor is added, a sparse noise reduction self-encoder (SDAE) is obtained, and then the cost function primitive can be expressed as:
JSDAE＝JAE+γPN
wherein, gamma represents the weight of the sparse penalty factor; j. the design is a squareAEIs a stacked autoencoder cost function; PN represents a sparse term penalty factor.
S402) pre-training the data samples;
during the pre-training process of the sparse autoencoder, the connection weight and the bias need to be updated according to the following formula
Figure BDA0003482687960000101
Figure BDA0003482687960000102
In the formula: ε represents the learning rate, JSDAERepresenting a sparse noise reduction autoencoder cost function, Wij(l) Representing the weight between the ith and jth neurons of the l-th layer, bi(l) Representing the offset between the ith layer and the ith neuron,
Figure BDA0003482687960000103
the difference is indicated.
S403) taking the hidden layer output value of the previous sparse noise reduction self-encoder as the input value of the next sparse noise reduction self-encoder, and repeating the previous step by adopting a layer-by-layer greedy algorithm until all self-encoders are trained completely.
S404) adding a softmax classifier to the sparse noise reduction self-encoder for classification.
S405) updating the weight and the bias in each iteration by using a BP back propagation algorithm, and finely adjusting the whole depth network parameters.
S406) testing the algorithm classification accuracy by using the test set.
S407) completing the deep neural network model.
In one specific embodiment of step S406), the classification accuracy is tested using the test set to see if the expected accuracy is achieved, and if not, the step S405) is repeated to fine tune the deep neural network model until the expected accuracy is achieved, and then the trained model is obtained. Wherein the expected accuracy may be 90%.
An embodiment of the present invention further provides a device for implementing the above-mentioned method for intelligently diagnosing an important service water pump, as shown in fig. 5, where the method includes:
and the acceleration sensor 501 is used for acquiring vibration signals of different fault types.
The data acquisition unit 502 is used for isolating and normalizing input voltage and converting analog input quantity into digital quantity required by a microcomputer system so as to interface with a CPU.
The signal processing unit 503 is configured to perform wavelet threshold denoising preprocessing on the acquired signal.
And the intelligent diagnosis unit 504 is used for establishing a fault game model and a deep neural network model, intelligently identifying fault signals and automatically judging fault types by adopting the deep neural network.
And the display unit 505 is used for labeling the classified fault types on a human-computer interaction interface, and when a fault occurs, early warning the corresponding fault types in time.
And a storage unit 506 for storing data instantly and reading and writing data.
And a reset unit 507, configured to reset the current CPU and the running data to start after the current CPU and the running data are cleared.
In the embodiment of the present invention, the data acquisition unit 502 is connected to the LabVIEW measurement system through an acceleration sensor, and the intelligent diagnosis unit 504 is completed by reading and processing data through the LabVIEW.
The device and method embodiments in the embodiments of the invention are based on the same inventive concept.
In the embodiment of the present invention, units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, and may be distributed over a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (11)

1. An intelligent diagnosis method for important service water pumps is characterized by comprising the following steps:
(S1) collecting vibration signals of important service water pumps under different working conditions to obtain sample points containing the vibration signals in different fault states;
(S2) performing wavelet threshold denoising pre-processing on the sample points;
(S3) building a fault game model to provide an interactive environment for fault diagnosis agents to observe, act on and obtain rewards;
(S4) reducing dimensions and extracting features of a plurality of hidden layers by stacking self-coding neural networks, and optimizing initial parameters by using a BP neural network to obtain a deep neural network model with the functions of feature extraction and pattern recognition;
(S5) outputting the diagnosis result.
2. The intelligent diagnosis method for essential service water pumps as claimed in claim 1, wherein in the step (S1), vibration signals in different fault states are collected through a measurement system consisting of an acceleration sensor and LabVIEW software.
3. The intelligent diagnosis method for essential service water pumps according to claim 1 or 2, wherein in the step (S1), the causes of the failure include: the impeller is characterized in that severe abrasion and cavitation occur at the position of the opening ring, the center of the rotor is not coincident with the center of the volute, the blades are not properly aligned with the diffuser, the design of an inlet pipeline and an outlet pipeline is unreasonable, the inlet pipeline and the outlet pipeline are not provided with straight pipe sections or the straight pipe sections are too short to cause front turbulence of fluid entering the impeller, and the blades are not operated under the designed working condition and are easy to generate turbulence, desludging, impact and random uneven impact of cavitation on the blades.
4. The method for intelligently diagnosing the essential service water pump as claimed in claim 1, wherein the method for performing the denoising preprocessing on the sample points in the step (S2) comprises the steps of:
(S201) denoising the noise data by improving a threshold function, and selecting an improved threshold th as follows:
Figure FDA0003482687950000021
wherein σ2And the mean is the median of the absolute value of the high-frequency wavelet coefficient, n is the signal length, j is the decomposition layer number of the wavelet, and beta is a control factor.
(S202) extracting wavelet packet energy of de-noising data by utilizing wavelet packet transformation;
(S203) carrying out wavelet packet decomposition on the vibration signal in the fault state;
(S204) the wavelet packet energy of each node after decomposition is obtained according to the following formula:
Ejn＝1M∑t＝1M((Xjn(t))2
wherein, Xjn(t) represents a time-domain vibration signal, t represents time, and M is a node XjnThe number of samples in (c).
5. The intelligent diagnosis method for essential service water pumps according to claim 4, wherein the wavelet basis selected in the step (S203) is db3, and the number of decomposition layers is 4.
6. The intelligent diagnosis method for the essential service water pump according to claim 1, wherein the specific process of the step (S3) comprises:
(S301) generating a fault diagnosis problem;
(S302) accepting the diagnostic question via the agent and giving a result feedback;
(S303) judging whether the diagnosis result is correct for the game model, if the correct total return is added with 1, outputting a game return value, otherwise, subtracting 1 from the total return, and returning to (S301) to regenerate the fault diagnosis problem.
7. The intelligent diagnosis method for the essential service water pump according to claim 1, wherein the specific process of the step (S4) comprises:
(S401) adding damage noise to the preprocessed data, and adding noise reduction limitation and sparsity limitation to an automatic encoder to obtain a sparse noise reduction self-encoder;
(S402) pre-training the sparse noise reduction self-encoder;
(S403) taking the hidden layer output value of the previous sparse noise reduction self-encoder as the input value of the next sparse noise reduction self-encoder, and repeating the step (S402) by adopting a layer-by-layer greedy algorithm until all the sparse noise reduction self-encoders are trained;
(S404) adding a softmax classifier to the sparse noise reduction self-encoder for classification;
(S405) updating the weight and the bias in each iteration by using a BP back propagation algorithm, and finely adjusting the whole depth network parameter;
(S406) testing the algorithm classification accuracy by using the test set;
(S407) completing the deep neural network model.
8. The method according to claim 7, wherein in step (S401), the noise reduction limitation is modified by using a mask noise, that is, the input data is randomly set to 0, the modified input vector is recorded as x', and the activation expression is:
h＝f(x')＝Sf(Wx′+b)
in the formula: x' is the modified input vector; sfRepresenting a non-linear activation function; w is an encoder weight matrix; b is the encoder bias;
sparsity constraints are achieved by controlling the average activation of neurons in the hidden layer, assuming aj(x) Representing the jth activation unit of the hidden layer, the average activation amount of the jth unit of the hidden layer is:
Figure FDA0003482687950000031
in the formula: n denotes the total number of samples, and j denotes the hidden layer unit.
To ensure pjNo deviation of rho, a sparse term penalty factor needs to be added into the cost function
Figure FDA0003482687950000032
In the formula: s2Representing the number of hidden layer neurons; rho is a sparsity parameter; KL is Kullback-Leibler (induced sparsity) divergence; rhojRepresenting the average activation amount of the jth unit of the hidden layer;
after the noise reduction limitation is performed on the stacked self-encoder AE and the sparse penalty factor is added, a sparse noise reduction self-encoder SDAE is obtained, and then the original cost function can be expressed as:
JSDAE＝JAE+γPN
wherein, gamma represents the weight of the sparse penalty factor; j. the design is a squareAEIs a stacked autoencoder cost function; PN represents a sparse term penalty factor.
9. The method for intelligently diagnosing a critical service water pump according to claim 7 or 8, wherein in the step (S402), the sparse noise reduction self-encoder updates the connection weight and the bias according to the following formula in the pre-training process:
Figure FDA0003482687950000041
Figure FDA0003482687950000042
in the formula: ε represents the learning rate, JSDAERepresenting a sparse noise reduction autoencoder cost function, Wij(l) Representing the weight between the ith and jth neurons of the l-th layer, bi(l) Representing the offset between the ith layer and the ith neuron,
Figure FDA0003482687950000043
the difference is indicated.
10. The intelligent diagnosis method for the essential service water pump according to claim 1, wherein the specific process of the step (S5) comprises: and the display unit in the computer equipment gives an early warning on the fault type corresponding to the vibration signal and stores the fault type in the storage unit in time.
11. An apparatus for implementing the intelligent diagnosis method for a service water pump according to any one of claims 1 to 10, comprising:
the acceleration sensor is used for acquiring vibration signals of different fault types;
the data acquisition unit is used for isolating and normalizing input voltage and converting analog input quantity into digital quantity required by a microcomputer system so as to interface with a CPU (central processing unit);
the signal processing unit is used for carrying out wavelet threshold denoising pretreatment on the acquired signals;
the intelligent diagnosis unit is used for establishing a fault game model and a deep neural network model, intelligently identifying fault signals and automatically judging fault types by adopting the deep neural network;
the display unit is used for labeling the classified fault types on a human-computer interaction interface, and warning the corresponding fault types in time after the fault occurs;
the storage unit is used for instantly storing data and reading and writing the data;
and the reset unit is used for starting the current CPU and the running data after being cleared under the condition of no power failure.
CN202210072680.2A 2022-01-21 2022-01-21 Intelligent diagnosis method and device for important service water pump Pending CN114548154A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210072680.2A CN114548154A (en) 2022-01-21 2022-01-21 Intelligent diagnosis method and device for important service water pump

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210072680.2A CN114548154A (en) 2022-01-21 2022-01-21 Intelligent diagnosis method and device for important service water pump

Publications (1)

Publication Number Publication Date
CN114548154A true CN114548154A (en) 2022-05-27

Family

ID=81672579

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210072680.2A Pending CN114548154A (en) 2022-01-21 2022-01-21 Intelligent diagnosis method and device for important service water pump

Country Status (1)

Country Link
CN (1) CN114548154A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115062553A (en) * 2022-08-17 2022-09-16 浪潮通用软件有限公司 Water pump working condition degradation detection method, equipment and medium based on multi-model fusion
CN115222164A (en) * 2022-09-20 2022-10-21 国能大渡河大数据服务有限公司 Water pump fault prediction method and system based on empirical coupling function

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115062553A (en) * 2022-08-17 2022-09-16 浪潮通用软件有限公司 Water pump working condition degradation detection method, equipment and medium based on multi-model fusion
CN115062553B (en) * 2022-08-17 2022-11-25 浪潮通用软件有限公司 Water pump working condition degradation detection method, equipment and medium based on multi-model fusion
CN115222164A (en) * 2022-09-20 2022-10-21 国能大渡河大数据服务有限公司 Water pump fault prediction method and system based on empirical coupling function

Similar Documents

Publication Publication Date Title
CN110728654B (en) Automatic pipeline detection and classification method based on deep residual error neural network
CN111830408B (en) Motor fault diagnosis system and method based on edge calculation and deep learning
CN113884290B (en) Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network
CN109308522B (en) GIS fault prediction method based on recurrent neural network
CN111323220A (en) Fault diagnosis method and system for gearbox of wind driven generator
CN114548154A (en) Intelligent diagnosis method and device for important service water pump
CN111507884A (en) Self-adaptive image steganalysis method and system based on deep convolutional neural network
CN111783362B (en) Method and system for determining residual service life of electric gate valve
CN113935406A (en) Mechanical equipment unsupervised fault diagnosis method based on counter current model
CN112347571B (en) Rolling bearing residual life prediction method considering model and data uncertainty
CN112611584B (en) Fatigue failure detection method, device, equipment and medium for wind generating set
CN112132102A (en) Intelligent fault diagnosis method combining deep neural network with artificial bee colony optimization
CN114037478A (en) Advertisement abnormal flow detection method and system, electronic equipment and readable storage medium
CN115587543A (en) Federal learning and LSTM-based tool residual life prediction method and system
CN115290326A (en) Rolling bearing fault intelligent diagnosis method
CN114441173A (en) Rolling bearing fault diagnosis method based on improved depth residual shrinkage network
CN115791174B (en) Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium
CN117037841A (en) Acoustic signal hierarchical cavitation intensity identification method based on hierarchical transition network
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
CN113409213B (en) Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump
CN115540935A (en) Novel method for diagnosing equipment fault of ship-protecting ballast water system
CN112560252A (en) Prediction method for residual life of aircraft engine
Zheng et al. An unsupervised transfer learning method based on SOCNN and FBNN and its application on bearing fault diagnosis
CN116701948B (en) Pipeline fault diagnosis method and system, storage medium and pipeline fault diagnosis equipment
CN112765034B (en) Software defect prediction method based on neural network

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
CB03 Change of inventor or designer information

Inventor after: Huang Qian

Inventor after: Zhang Rongyong

Inventor after: Zhi Yifan

Inventor after: Li Qi

Inventor after: Zhang Wenjie

Inventor after: Dai Li

Inventor after: Li Na

Inventor before: Zhang Rongyong

Inventor before: Zhi Yifan

Inventor before: Li Qi

Inventor before: Zhang Wenjie

Inventor before: Dai Li

Inventor before: Li Na

Inventor before: Huang Qian

CB03 Change of inventor or designer information