CN113361016A - Ship auxiliary boiler fault diagnosis method and device - Google Patents

Ship auxiliary boiler fault diagnosis method and device Download PDF

Info

Publication number
CN113361016A
CN113361016A CN202110736853.1A CN202110736853A CN113361016A CN 113361016 A CN113361016 A CN 113361016A CN 202110736853 A CN202110736853 A CN 202110736853A CN 113361016 A CN113361016 A CN 113361016A
Authority
CN
China
Prior art keywords
neural network
operation data
fault diagnosis
model
auxiliary boiler
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
CN202110736853.1A
Other languages
Chinese (zh)
Inventor
甘辉兵
王世威
胡国彤
鲁道毅
张均东
何治斌
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.)
Dalian Maritime University
Original Assignee
Dalian Maritime 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 Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN202110736853.1A priority Critical patent/CN113361016A/en
Publication of CN113361016A publication Critical patent/CN113361016A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/15—Vehicle, aircraft or watercraft design
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/25—Design optimisation, verification or simulation using particle-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • 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
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Automation & Control Theory (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention provides a ship auxiliary boiler fault diagnosis method and device. The invention comprises the following steps: acquiring the operation data of the auxiliary boiler of the ship, and verifying the correlation coefficient of the operation data of the auxiliary boiler to obtain a correlation coefficient table; normalizing the auxiliary boiler operation data, and adding random noise to obtain first operation data; performing principal component analysis on the first operation data to generate second operation data; and carrying out fault diagnosis on the second operation data through a trained fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis model is a mixed neural network model consisting of a BP neural network and an SOM neural network. The invention adopts a normalization method to eliminate the influence between dimensions and magnitude of different diagnostic parameters, and adopts a noise processing method to simulate the real ship data acquisition environment, thereby improving the authenticity. The principal component analysis method is adopted to eliminate the influence of redundant data characteristics and data correlation, simplify the structure of the model and improve the convergence efficiency of the model.

Description

Ship auxiliary boiler fault diagnosis method and device
Technical Field
The invention relates to the technical field of ship turbine engineering, in particular to a ship auxiliary boiler fault diagnosis method and device.
Background
The auxiliary boiler for the ship is mainly used for generating saturated steam to heat fuel oil, preheating mechanical equipment and meeting the requirements of various life and work on the ship, and is an important component for normal operation of the ship. However, due to the complexity of the structure of the marine auxiliary boiler and the badness of the operating environment, there is a great demand for the reliability of the marine auxiliary boiler. When the auxiliary boiler of the ship breaks down, if the fault cannot be found in time or the fault reason cannot be found out, economic loss of ship operation is caused, and the life and property safety of ship management personnel is threatened. Therefore, in order to meet the new requirements of ship intellectualization and guarantee the life and property safety of marine personnel, a fault diagnosis method suitable for a ship auxiliary boiler needs to be researched.
With the continuous rise of artificial intelligence, the neural network is widely applied to fault diagnosis by virtue of strong nonlinear fitting and learning capability of the neural network. The method for researching and researching the fault diagnosis of the ship auxiliary boiler based on the neural network is helpful for ensuring the safety and the economy of the ship operation. Aiming at the problem of fault diagnosis of the ship auxiliary boiler, the fault diagnosis of the ship auxiliary boiler is rarely researched at present, most researches relate to power station boilers, and the complexity of the running environment of the ship auxiliary boiler and the difficulty of data acquisition bring great challenges to the fault diagnosis. A plurality of researchers apply the multidimensional BP neural network to boiler fault diagnosis modeling, boiler fault diagnosis can be rapidly and effectively carried out, meanwhile, a plurality of researchers optimize the neural network by utilizing the optimizing capability of the intelligent optimization algorithm, and a good diagnosis effect is achieved. Therefore, a single neural network cannot meet the requirements of complexity and diversity of fault diagnosis of the auxiliary boiler of the ship under the current situation. Therefore, it has become a trend to use a hybrid neural network or an optimized neural network for pattern recognition and fault diagnosis.
Disclosure of Invention
According to the conventional ship auxiliary boiler provided by the invention, under the complex system working condition and the severe operation environment, the traditional fault diagnosis cannot find the fault point in time only by depending on experience and the problem that the ship is easy to generate alarm fatigue is solved, so that the ship auxiliary boiler fault diagnosis method and the device are provided. The technical means adopted by the invention are as follows:
a ship auxiliary boiler fault diagnosis method comprises the following steps:
step 1, acquiring ship auxiliary boiler operation data, performing correlation coefficient verification on the auxiliary boiler operation data, judging whether each fault diagnosis characteristic parameter is at least related to one fault, and obtaining a correlation coefficient table;
step 2, carrying out normalization processing on the auxiliary boiler operation data, eliminating the influence of magnitude and dimension of each diagnosis parameter, adding random noise, and simulating a real ship acquisition environment to obtain first operation data;
step 3, performing principal component analysis on the first operation data, reducing information redundancy, simplifying a subsequent neural network model structure, and generating second operation data, wherein the dimensionality of the second operation data is smaller than that of the first operation data;
step 4, performing fault diagnosis on the second operation data through a trained fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis model is a hybrid neural network model composed of a BP neural network and an SOM neural network, the fault diagnosis model is used for classifying the second operation data, and the fault diagnosis result comprises a fault name corresponding to the category to which the second operation data belongs;
in the step 4, the initial connection weight and the threshold of the hybrid neural network model are optimized by adopting a particle swarm optimization algorithm so as to improve the diagnosis accuracy, the diagnosis precision and the convergence speed.
Further, the hybrid neural network diagnostic model includes:
inputting second operation data obtained by noise adding, normalization and principal component analysis of the collected operation data of the ship auxiliary boiler into an SOM neural network serving as a primary neural network to obtain winning neurons;
and the output of the primary neural network is used as a one-dimensional vector and is input into the secondary BP neural network together with second operation data to obtain a diagnosis result.
Further, the initial connection weight and node threshold optimizing process of the hybrid neural network comprises the following steps:
forming particles in the particle swarm by using the initial connection weight and the node threshold to be optimized of the hybrid neural network in a vector coding mode;
taking the mean square error of the hybrid neural network as a fitness function of the particles, wherein the mean square error of the hybrid neural network is the mean square error of the actual output and the expected output when the hybrid neural network diagnosis model carries out fault identification on historical data of the ship auxiliary boiler system;
and optimizing the hybrid neural network diagnosis model based on a particle swarm optimization algorithm, optimizing a connection weight and a node threshold according to a fitness function, and storing the optimized excellent particles.
The invention also discloses a ship auxiliary boiler fault diagnosis device, which comprises:
the data acquisition unit is used for acquiring running data of each state of the ship auxiliary boiler to obtain the running data of the auxiliary boiler;
the processing unit is used for carrying out normalization processing on the auxiliary boiler operation data, adding random noise, eliminating the influence of the magnitude of each diagnosis parameter and dimension, simulating a real ship acquisition environment and generating first operation data;
the dimensionality reduction unit is used for performing principal component analysis on the first operation data to generate second operation data, and the dimensionality of the second operation data is smaller than that of the first operation data;
the fault diagnosis unit is used for identifying the fault state of the second operation data through the trained fault diagnosis model identification to obtain a fault diagnosis result; the fault diagnosis model is used for classifying the second operation data to obtain fault names corresponding to the second operation data types, and the fault diagnosis result comprises the fault names corresponding to the second operation data types.
Further, the fault diagnosis model is a hybrid neural network model, and an initial connection weight and a node threshold of the hybrid neural network model are obtained through a particle swarm optimization algorithm.
Further, the optimization process of the initial connection weight and the node threshold of the hybrid neural network model comprises the following steps:
forming particles by the initial connection weight and the node threshold of the hybrid neural network model;
taking the training mean square error of the hybrid neural network model as a fitness function of the particles, wherein the mean square error is the mean square error of the diagnosis model for fault recognition of historical data of the ship auxiliary boiler system;
and optimizing the initial connection weight and the threshold according to a fitness function by the particle swarm optimized hybrid neural network diagnostic model, and storing the excellent particles as the initial connection weight and the threshold of the hybrid neural network diagnostic model.
The invention has the following advantages:
1. the invention adopts a normalization method to eliminate the influence between dimensions and magnitude of different diagnostic parameters, and adopts a noise processing method to simulate the real ship data acquisition environment, thereby improving the authenticity. The data are normalized, and the influence of different dimensions and magnitude of various diagnostic parameters on the diagnostic result is eliminated. The principal component analysis method is adopted to eliminate the influence of redundant data characteristics and data correlation, simplify the structure of the model and improve the convergence efficiency of the model.
2. The invention adopts the BP neural network and the SOM neural network to form the hybrid neural network, thereby making up the limitation of a single neural network. And a particle swarm optimization algorithm is introduced to optimize the SOM-BP hybrid neural network diagnosis model, so that the accuracy, the convergence efficiency and the diagnosis precision of fault diagnosis are improved. The established ship auxiliary boiler fault diagnosis model based on the particle swarm optimization hybrid neural network improves the convergence speed, the diagnosis precision and the diagnosis accuracy of the ship auxiliary boiler fault diagnosis.
Based on the reasons, the invention can be widely popularized in the field of ship turbine engineering.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a ship auxiliary boiler fault diagnosis method of the invention.
FIG. 2 is a flow chart of a particle swarm optimization hybrid neural network model according to the present invention.
FIG. 3 is a diagram illustrating the contribution rate of each principal component and the cumulative contribution rate in the embodiment of the present invention.
FIG. 4 is a graph of error as a function of the number of hidden layer nodes in the embodiment of the present invention.
FIG. 5 is a graph of the training error of the unoptimized hybrid neural network in an embodiment of the present invention.
Fig. 6 is a particle swarm optimization fitness curve in the embodiment of the invention.
FIG. 7 is a particle swarm optimization neural network training error curve in an embodiment of the present invention.
FIG. 8 is a diagram of the results of the particle swarm optimization model diagnostic tests of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, an embodiment of the invention discloses a ship auxiliary boiler fault diagnosis method, which includes the following steps:
step 1, acquiring ship auxiliary boiler operation data, verifying a correlation coefficient of the auxiliary boiler operation data, judging whether each fault diagnosis characteristic parameter is at least related to one fault, and obtaining a correlation coefficient table;
step 2, normalizing the auxiliary boiler operation data, adding random noise, eliminating the influence of magnitude and dimension of each diagnosis parameter, simulating a real ship acquisition environment to obtain first operation data, wherein 1.5% of random noise is added in the embodiment;
step 3, performing principal component analysis on the first operation data, reducing information redundancy, simplifying a subsequent neural network model structure, and generating second operation data, wherein the dimensionality of the second operation data is smaller than that of the first operation data;
and 4, carrying out fault diagnosis on the second operation data through the trained fault diagnosis model to obtain a fault diagnosis result so as to improve the diagnosis accuracy, the diagnosis precision and the convergence speed. The fault diagnosis model is a hybrid neural network model composed of a BP neural network and an SOM neural network, the fault diagnosis model is used for classifying the second operation data, and the fault diagnosis result comprises a fault name corresponding to the category to which the second operation data belongs;
in the step 4, the initial connection weight and the threshold of the hybrid neural network model are optimized by adopting a particle swarm optimization algorithm.
The hybrid neural network diagnostic model includes:
inputting second operation data obtained after noise addition, normalization and principal component analysis into an SOM neural network serving as a primary neural network to obtain winning neurons;
and the output of the primary neural network is used as a one-dimensional vector and is input into the secondary BP neural network together with second operation data to obtain a diagnosis result.
The initial connection weight and node threshold optimizing process of the hybrid neural network comprises the following steps:
forming particles in the particle swarm by using the initial connection weight and the node threshold to be optimized of the hybrid neural network in a vector coding mode;
taking the diagnosis mean square error of the hybrid neural network as a fitness function of the particles, wherein the diagnosis mean square error is the mean square error of the hybrid neural network diagnosis model for carrying out fault identification on historical data of the ship auxiliary boiler system;
and optimizing the hybrid neural network diagnosis model based on a particle swarm optimization algorithm, optimizing a connection weight and a node threshold according to a fitness function, and storing the optimized excellent particles.
The optimization process is shown in fig. 2 and includes:
a. initializing particles, encoding the initial connection weight and the node threshold of the hybrid neural network model, and taking the encoded initial connection weight and the node threshold as the particles of the particle swarm algorithm.
b. And (3) establishing a fitness function, and taking the mean square error of the training data as the fitness function to obtain a stable diagnosis model with high precision and high convergence speed.
c. And (3) establishing a fault diagnosis model, wherein 7 new characteristic data sets obtained after normalization, noise addition and principal component analysis are used as the input of the hybrid neural network model.
d. The particles after particle swarm optimization are used as initial connection weights and node thresholds of the hybrid neural network model, so that the diagnosis accuracy, the diagnosis precision and the convergence efficiency are improved.
The embodiment of the invention also discloses a ship auxiliary boiler fault diagnosis device, which comprises:
the data acquisition unit is used for acquiring running data of each state of the ship auxiliary boiler to obtain the running data of the auxiliary boiler;
the processing unit is used for carrying out normalization processing on the auxiliary boiler operation data, adding random noise, eliminating the influence of the magnitude of each diagnosis parameter and dimension, simulating a real ship acquisition environment and generating first operation data;
the dimensionality reduction unit is used for performing principal component analysis on the first operation data to generate second operation data, and the dimensionality of the second operation data is smaller than that of the first operation data;
the fault diagnosis unit is used for identifying the second operation data through the trained fault diagnosis model to obtain a fault diagnosis result; the fault diagnosis model is used for classifying the second operation data to obtain fault names corresponding to the second operation data types, and the fault diagnosis result comprises the fault names corresponding to the second operation data types.
The fault diagnosis model is a hybrid neural network model, and an initial connection weight and a node threshold of the hybrid neural network model are obtained through a particle swarm optimization algorithm.
The optimization process of the initial connection weight and the node threshold of the hybrid neural network model comprises the following steps:
forming particles by the initial connection weight and the node threshold of the hybrid neural network model;
taking the training mean square error of the hybrid neural network model as a fitness function of the particles, wherein the mean square error is the mean square error of the diagnosis model for fault recognition of historical data of the ship auxiliary boiler system;
and optimizing the initial connection weight and the threshold according to a fitness function by the particle swarm optimized hybrid neural network diagnostic model, and storing the excellent particles as the initial connection weight and the threshold of the hybrid neural network diagnostic model.
Example 1
In the embodiment, in the data acquisition step, the ship auxiliary boiler of the DMS-VLCC type turbine simulator developed by university of maritime affairs is used as a data source, the simulation model data of the platform is compared with the design value, the error of the platform is within 1.5%, the precision requirement is met, and the actual boiler state can be truly reflected. Therefore, fault experiments are carried out on the simulation experiment platform, and fault sample data of the ship auxiliary boiler in 10 states are collected. In order to eliminate the influence of different dimensions and magnitude of each fault characteristic parameter in the fault diagnosis of the ship auxiliary boiler, the collected auxiliary boiler operation data is normalized, and the diagnosis parameters are mapped to a [0-1] interval, so that the problem of low convergence speed of neural network training caused by different data types and value ranges is solved. Due to the special operating environment of the vessel, data acquisition is susceptible to noise, high temperatures and vibration. 1.5% of random noise is added into the sample data set, so that the authenticity of fault diagnosis is improved, and first operation data are generated. It is to be noted that in the principal component analysis, the normalization process is not required.
In the embodiment, 9 typical fault states and normal states are selected for carrying out fault diagnosis experiments. I.e. class 10 state: normal state, pre-heater dirty and blocked, fuel supply pump abrasion, fan failure, ignition oil pump abrasion, main steam pipeline dirty and blocked, condenser dirty and blocked, heat exchange surface dirty, boiler feed pump failure and water circulation pump failure. As shown in table 1, the monitoring points for each selected diagnostic parameter are shown in table 2, and there are 26 diagnostic parameters in total.
1. Data source and preprocessing:
(1) sample data extraction
The acquisition experiment system comprises 10 states of experiment samples including a normal state and 9 fault states, the specific states and state numbers are shown in table 1, 26 fault diagnosis characteristic parameters are shown in table 2, 110 groups of samples are acquired in each state, part of original sample data sets are shown in table 3, 1100 groups of experiment samples are obtained, and the size of the experiment samples is 1100 × 26.
(2) Correlation analysis
And performing relevance analysis on the acquired sample data by using SPSS software, analyzing whether each diagnosis parameter is relevant to the fault, removing the characteristic with lower discrimination degree by analyzing the relevant characteristics of different characteristics, and reserving the characteristic with higher lower discrimination degree to obtain a relevant coefficient table 4. It can be seen that each characteristic parameter has a significant correlation with at least one fault, so that 26 characteristic parameters are reserved as input parameters of the neural network fault diagnosis model.
(3) Denoising and normalization processing of sample data
In order to simulate a real ship acquisition environment and eliminate the influence of dimensions and magnitude among different diagnosis parameters on a diagnosis effect, noise addition and normalization processing are carried out on sample data. Due to the special operating environment of the vessel, data acquisition is susceptible to noise, high temperatures and vibration. 1.5% of random noise is added into the sample data set, so that the authenticity of fault diagnosis is improved. The noise adding formula and the normalization formula of the sample data are as shown in formulas (1) and (2):
Figure BDA0003141946080000081
Figure BDA0003141946080000082
in the formula xnFor the purpose of the data after the noise addition,
Figure BDA0003141946080000083
average value of data for each type of operating state, xpFor normalized data, xrAs actual data, xmaxAnd xminThe maximum value and the minimum value in the characteristic parameter.
TABLE 1 diagnostic State numbering and State
Figure BDA0003141946080000084
Figure BDA0003141946080000091
TABLE 2 part failure diagnosis characteristic parameters of ship auxiliary boiler
Diagnostic parameter Unit of (symbol) Diagnostic parameter Unit of (symbol)
Temperature of main oil circuit ℃ Tz Main steam pipe flow m3/s Qv
Main oil path flow t/h Qz Main steam pipe pressure MPa Pv
Pressure of main oil circuit MPa Pz Main steam line temperature ℃ Tv
Temperature of boiler steam ℃ Tg Condenser hot water well temperature ℃ Tc
Boiler steam pressure bar Qg Air door opening degree ％ Dp
Boiler water level mm L Opening of water supply solenoid valve ％ Os
Fuel supply pump flow t/h Qp Outlet pressure of feed pump bar Ps
Fuel pump outlet pressure bar Po Outlet pressure of waste gas water circulating pump bar Pe
Fuel pump outlet temperature ℃ To Exhaust gas boiler outlet temperature ℃ Te
Flow of pre-heater steam kg/h Qh Flue gas temperature of boiler ℃ Tb
Flame receptor 0/1 Fl Pressure of boiler flue gas bar Pb
Maximum pressure of ignition oil pump MPa Pi Boiler flue gas flow kg/h Qb
Inlet flow of exhaust gas boiler kg/h Qe Inlet temperature of exhaust gas boiler ℃ Ti
Table 3 partial original sample data set
Figure BDA0003141946080000092
Table 4 correlation coefficient table
Figure BDA0003141946080000101
2. Data dimension reduction
The dimension reduction is carried out on the 26-dimensional characteristic parameters by adopting a principal component analysis method, so that the time of model diagnosis is reduced, and the efficiency of the model diagnosis is improved. The SPSS software was used to perform principal component analysis to obtain the variance contribution ratios of the components in Table 5 and the principal component coefficient matrix in Table 6. As shown in fig. 3, the cumulative variance contribution rate of the first 7 principal components reaches 96.75%, which shows that the 7 principal components can represent 96% of the original 26 parameters of information, which can represent sample data. Table 6 shows the calculated principal component coefficient matrix. The principal component solving formula is as follows:
Figure BDA0003141946080000102
wherein Y isiIs the ith main component, xjIs the jth parameter of the sample dataNumber, ZijIs a principal component matrix. In summary, after principal component analysis is performed on 26 parameters in the original sample data, 26 independent new components are converted. The first 7 components may already represent features of the entire sample data. Therefore, only the first 7 characteristic data in the table 5 are needed to be used for fault diagnosis research, so that data redundancy is reduced, and the diagnosis efficiency of the model is improved.
TABLE 5 variance contribution ratio of each component
Composition (I) Characteristic value lambda Variance contribution ratio% Cumulative variance contribution%
1 9.985767 38.40679 38.40679
2 4.868777 18.72606 57.13286
3 4.142071 15.93104 73.0639
4 2.428899 9.341919 82.40582
5 1.664231 6.40089 88.80671
6 1.223049 4.704036 93.51075
7 0.842663 3.241012 96.75176
TABLE 6 principal component coefficient matrix
Figure BDA0003141946080000103
Figure BDA0003141946080000111
3. Establishing a hybrid neural network diagnostic model
1100 × 26 sample data are collected, and 7 feature data after principal component analysis are taken as a data set, wherein the data set comprises 1000 training data (100 pieces in each state) and 100 testing data (10 pieces in each state). And (3) utilizing MATLAB coding to realize a hybrid neural network diagnostic model. The output layer of the hybrid neural network is determined by the dimensionality of an input sample, and an original sample is converted into a new 7-dimensional data characteristic after principal component analysis, so that the number of nodes of the input layer is 7; the design of the competition layer is the number of neurons of the competition layer and the topological structure, the excessive number of nodes can cause 'dead nodes' to appear, or the excessive number of nodes can cause samples with similar data and different categories to be gathered together, in order to avoid the problems, the experiments are respectively provided with the competition layers of 4 x 4, 4 x 5, 5 x 5 and 5 x 6, the diagnosis effect is optimal when the competition layer is designed to be 5 x 5 in the experiments, and the hexagonal topological structure is adopted in consideration of the actual needs of fault diagnosis and in order to more visually see the classification result; the number of neuron nodes of the secondary BP input layer is 8, namely 7 fault diagnosis characteristic parameters and 1 SOM neural network output value; the hidden layer design comprises the number of hidden layer layers and the number of nodes, generally, a BP neural network of one hidden layer can realize nonlinear mapping, in order to simplify the structure of a model as much as possible and reduce the network training time, the single hidden layer design is adopted in the text, meanwhile, the reasonable interval of the number of the hidden layer nodes is determined to be [3,19] according to an empirical formula, experiments are carried out on the neural networks of the hidden layer with different numbers, the mean square error and the percentage of classification errors are used as evaluation indexes to obtain a graph 4 of the error change of the number of the hidden layer nodes, when the number of the hidden layer nodes is 12, the diagnosis performance is optimal, and therefore, the number of the hidden layer nodes is set to be 12; the output layer design adopts 9-dimensional vectors to represent 10 diagnosis states, and the number of the neuron nodes is 9.
The structure of the hybrid neural network model is determined to be 7-5 multiplied by 5-8-12-9 through experimental tests, and the hybrid neural network model consists of an input layer, a competition layer, a hidden layer and an output layer. The principle is that input data is input through an input layer and then distributed to different competition layer neurons, namely, the preliminary classification of a sample is realized, and the classification result of the competition layer is used as the input of a secondary (BP) neural network, which is equivalent to the increase of the input dimension of the secondary neural network. The hybrid neural network model can effectively reduce the training time of the BP neural network, and the convergence speed is higher. Meanwhile, the SOM neural network does not need a large amount of data, so that the hybrid neural network does not need a large amount of data, and the actual requirement of fault diagnosis of the ship auxiliary boiler is met.
And taking the output of the SOM neural network diagnosis model as the input of the secondary BP neural network, and judging the fault type through expected output. The 10 states corresponding to F0-F9 are represented by 9-dimensional vectors. As shown in table 7. Meanwhile, by combining experiments and documents, the training method of the secondary BP neural network is determined to be an L-M algorithm, the learning rate is set to be 0.01, the logsig function is selected by the hidden layer transfer function through the experiments, and the purelin function is selected by the output layer transfer function.
TABLE 7 expected output Table
Status of state (symbol) Desired output
Normal state F0 0 0 0 0 0 0 0 0 0
Filth blockage of preheater F1 0 0 0 0 0 0 0 0 1
Wear of fuel supply pump F2 0 0 0 0 0 0 0 1 0
Ignition oil pump wear F3 0 0 0 0 0 0 1 0 0
Fan failure F4 0 0 0 0 0 1 0 0 0
Dirty block of steam pipeline F5 0 0 0 0 1 0 0 0 0
Dirty block of condenser F6 0 0 0 1 0 0 0 0 0
Dirty heat exchange surface of boiler F7 0 0 1 0 0 0 0 0 0
Failure of boiler feed pump F8 0 1 0 0 0 0 0 0 0
Water circulation pump failure F9 1 0 0 0 0 0 0 0 0
4. Particle swarm optimization diagnosis model
Because the clustering efficiency and the clustering precision of the hybrid neural network model are easily influenced by improper setting of the initial connection weight and the threshold, the model diagnosis accuracy is further improved and the diagnosis time is reduced. And the initial connection weight and the threshold of the hybrid neural network model are optimized by adopting a particle swarm optimization algorithm, so that a ship auxiliary boiler fault diagnosis model is established, and a better diagnosis effect is obtained.
5. Experiment and analysis of results
(1) Hybrid neural network model diagnostic result analysis
And (5) training the 1000 processed sample sets as the input of the hybrid neural network model to obtain a training error curve chart 5. It can be seen that when the number of training iterations reaches 107, the model converges to the expected error, i.e. the training is complete. After model training, 100 test samples are input into the mixed neural network diagnosis model for diagnosis, and the average diagnosis accuracy of the obtained diagnosis model is 89%. The diagnosis result shows that the hybrid neural network diagnosis model can realize fault diagnosis to a certain extent, but the effect is not ideal and still needs to be optimized.
(2) Particle swarm optimization hybrid neural network diagnosis result analysis
The parameter setting of the particle swarm is determined according to experience and experiment, the inertia weight w is dynamic inertia weight, and a learning factor c1And c2Respectively set to 1.5 and 1.5, when the iteration reaches 100 times in the training process, the algorithm converges to find the optimal solution, so the iteration time TmaxSet to 100 times. The inertia weight of the particle swarm algorithm adopts a dynamic inertia weight adjustment mode wmax＝0.9，wminWhen the value is 0.4, the specific adjustment formula is as follows:
Figure BDA0003141946080000131
in the formula wmaxAnd wminRespectively representing the maximum and minimum inertial weight, TmaxAnd t is the maximum iteration number, and t is the current iteration number. The data are input into a neural network and subjected to a simulation experiment, and the optimization curve of the obtained particle swarm is shown in fig. 6. When the particle swarm algorithm iterates 93 times, the fitness value of the particle swarm algorithm reaches the lowest value, and the particle swarm algorithm shows that the optimal initial connection weight and node threshold of the hybrid neural network model are found. The initial connection weight and the node threshold value of the hybrid neural network model are stored, the initial connection weight and the node threshold value are assigned to the hybrid neural network, the ship auxiliary boiler fault diagnosis model is constructed and trained to obtain an optimal model, a training error curve is shown in FIG. 7, and it can be seen that when the training iteration number reaches 76 times, the model converges to an expected error, and the training is completed. The test data is input into the optimized neural network for diagnosis, and a diagram of the diagnosis result is obtained as shown in fig. 8.
The table 8 obtained through calculation is a diagnosis comparison table of the diagnosis model, and it can be seen from table 8 that the diagnosis accuracy of the particle swarm optimized hybrid neural network model is higher than that of the hybrid neural network model, the mean square error (generalization error) is smaller, and the recognition accuracy is higher. The particle swarm optimization algorithm has a good effect on improving the initial connection weight and adjusting the node threshold value, and is effective in fault diagnosis of the ship auxiliary boiler.
TABLE 8 hybrid neural network diagnostic results after particle swarm optimization
Diagnostic model Number of samples tested Number of wrong samples Mean square error Number of training sessions Accuracy of diagnosis
Hybrid neural network diagnostic model 100 11 0.0158 107 89％
Particle swarm optimization model 100 4 0.0108 76 96％
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (6)

1. A ship auxiliary boiler fault diagnosis method is characterized by comprising the following steps:
step 1, acquiring ship auxiliary boiler operation data, verifying a correlation coefficient of the auxiliary boiler operation data, judging whether each fault diagnosis characteristic parameter is at least related to one fault, and obtaining a correlation coefficient table;
step 2, normalizing the auxiliary boiler operation data, adding random noise, eliminating the influence of magnitude order and dimension of each diagnosis parameter, and simulating a real ship acquisition environment to obtain first operation data;
step 3, performing principal component analysis on the first operation data to generate second operation data, wherein the dimensionality of the second operation data is smaller than that of the first operation data;
step 4, performing fault diagnosis on the second operation data through a trained fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis model is a hybrid neural network model composed of a BP neural network and an SOM neural network, the fault diagnosis model is used for classifying the second operation data, and the fault diagnosis result comprises a fault name corresponding to the category to which the second operation data belongs;
in the step 4, the initial connection weight and the threshold of the hybrid neural network model are optimized by adopting a particle swarm optimization algorithm.
2. The ship auxiliary boiler fault diagnosis method according to claim 1, wherein the hybrid neural network diagnosis model comprises:
inputting second operation data obtained by noise adding, normalization and principal component analysis of the collected operation data of the ship auxiliary boiler into an SOM neural network serving as a primary neural network to obtain winning neurons;
and the output of the primary neural network is used as a one-dimensional vector and is input into the secondary BP neural network together with second operation data to obtain a diagnosis result.
3. The ship auxiliary boiler fault diagnosis method according to claim 1 or 2, wherein the initial connection weight and node threshold value optimizing process of the hybrid neural network comprises:
forming particles in the particle swarm by using the initial connection weight to be optimized and the node threshold of the hybrid neural network in a vector coding mode;
taking the mean square error of the hybrid neural network as a fitness function of the particles, wherein the mean square error of the hybrid neural network is the mean square error of the actual output and the expected output when the hybrid neural network diagnosis model carries out fault identification on historical data of the ship auxiliary boiler system;
and optimizing the hybrid neural network diagnosis model based on a particle swarm optimization algorithm, optimizing a connection weight and a node threshold according to a fitness function, and storing the optimized excellent particles.
4. A ship auxiliary boiler fault diagnosis device is characterized by comprising:
the data acquisition unit is used for acquiring running data of each state of the ship auxiliary boiler to obtain the running data of the auxiliary boiler;
the processing unit is used for carrying out normalization processing on the auxiliary boiler operation data, adding random noise, eliminating the influence of the magnitude of each diagnosis parameter and dimension, simulating a real ship acquisition environment and generating first operation data;
the dimensionality reduction unit is used for performing principal component analysis on the first operation data to generate second operation data, and the dimensionality of the second operation data is smaller than that of the first operation data;
the fault diagnosis unit is used for identifying the fault state of the second operation data through the trained fault diagnosis model to obtain a fault diagnosis result; the fault diagnosis model is used for classifying the second operation data to obtain fault names corresponding to the second operation data types, and the fault diagnosis result comprises the fault names corresponding to the second operation data types.
5. The ship auxiliary boiler fault diagnosis device according to claim 4, wherein the fault diagnosis model is a hybrid neural network model, and the initial connection weight and the node threshold of the hybrid neural network model are obtained through particle swarm optimization.
6. The ship auxiliary boiler fault diagnosis device according to claim 4, wherein the optimization process of the initial connection weight and the node threshold of the hybrid neural network model comprises:
forming particles by the initial connection weight and the node threshold of the hybrid neural network model;
taking the training mean square error of the hybrid neural network model as a fitness function of the particles, wherein the mean square error is the mean square error of the diagnosis model for fault recognition of historical data of the ship auxiliary boiler system;
and optimizing the initial connection weight and the threshold according to a fitness function by the particle swarm optimized hybrid neural network diagnostic model, and storing the excellent particles as the initial connection weight and the threshold of the hybrid neural network diagnostic model.
CN202110736853.1A 2021-06-30 2021-06-30 Ship auxiliary boiler fault diagnosis method and device Pending CN113361016A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110736853.1A CN113361016A (en) 2021-06-30 2021-06-30 Ship auxiliary boiler fault diagnosis method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110736853.1A CN113361016A (en) 2021-06-30 2021-06-30 Ship auxiliary boiler fault diagnosis method and device

Publications (1)

Publication Number Publication Date
CN113361016A true CN113361016A (en) 2021-09-07

Family

ID=77537386

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110736853.1A Pending CN113361016A (en) 2021-06-30 2021-06-30 Ship auxiliary boiler fault diagnosis method and device

Country Status (1)

Country Link
CN (1) CN113361016A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114048880A (en) * 2021-11-26 2022-02-15 江苏科技大学 Particle swarm optimization-based fault diagnosis method for opening valve

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111999088A (en) * 2020-08-29 2020-11-27 大连海事大学 Ship refrigeration system fault diagnosis method and device and storage medium
CN112036480A (en) * 2020-08-29 2020-12-04 大连海事大学 Ship refrigeration system fault diagnosis method and device and storage medium

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111999088A (en) * 2020-08-29 2020-11-27 大连海事大学 Ship refrigeration system fault diagnosis method and device and storage medium
CN112036480A (en) * 2020-08-29 2020-12-04 大连海事大学 Ship refrigeration system fault diagnosis method and device and storage medium

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
周晔锡: "基于SOM-BP串联神经网络的中速磁浮列车悬浮控制器故障诊断方法研究", 中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑), no. 01, pages 1 - 85 *
夏飞;罗志疆;张浩;彭道刚;张茜;唐依雯;: "混合神经网络在变压器故障诊断中的应用", 电子测量与仪器学报, no. 01 *
邹兰: "基于SOM-BP神经网络的盾构机故障诊断方法研究", 中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑, no. 12 *
高利民;许庆阳;李锋;杨吉;孟景辉;杨树忠;: "基于SOM-BP混合神经网络的道岔设备退化状态研究", 中国铁道科学, no. 03 *
高鹤元: "基于神经网络的船舶辅锅炉燃烧故障诊断研究", 中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑, no. 01, pages 1 - 92 *
高鹤元;甘辉兵;郑卓;杨远达;: "粒子群优化神经网络在船舶辅锅炉故障诊断中的应用", 计算机应用与软件, no. 08 *
黄群群;傅秀芬;许素霞;: "基于SOM-BP复合神经网络的多Agent入侵检测系统", 科技创新导报, no. 17 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114048880A (en) * 2021-11-26 2022-02-15 江苏科技大学 Particle swarm optimization-based fault diagnosis method for opening valve
CN114048880B (en) * 2021-11-26 2024-05-24 江苏科技大学 Particle swarm optimization-based opening valve fault diagnosis method

Similar Documents

Publication Publication Date Title
CN109740859A (en) Transformer condition evaluation and system based on Principal Component Analysis and support vector machines
CN104142254A (en) Fault diagnosis method for flow passage of steam turbine
Lu et al. An evaluation of engine faults diagnostics using artificial neural networks
CN113361016A (en) Ship auxiliary boiler fault diagnosis method and device
CN112464565B (en) Equipment fault early warning method integrating intelligent modeling and fuzzy rules
CN113640027A (en) Heat exchange unit fault diagnosis method and system based on edge calculation and neural network
CN112036480A (en) Ship refrigeration system fault diagnosis method and device and storage medium
Hou et al. A fault diagnosis model of marine diesel engine fuel oil supply system using PCA and optimized SVM
Tarassenko et al. Novelty detection in jet engines
CN116628592A (en) Dynamic equipment fault diagnosis method based on improved generation type countering network
CN112257335B (en) Oil-immersed transformer fault diagnosis method combining PNN and SVM
Li et al. Diagnosis for the refrigerant undercharge fault of chiller using deep belief network enhanced extreme learning machine
Rakhshani et al. Application of data mining on fault detection and prediction in boiler of power plant using artificial neural network
CN106289754B (en) A kind of fired power generating unit component capabilities variation determination method based on Statistical Control Technique
CN110702438B (en) Centrifugal water chilling unit and fresh air system combined fault diagnosis method
CN110289112A (en) Nuclear power unit health status diagnostic method based on step analysis and fuzzy evaluation
Wang et al. Convolutional domain adaptation network for fault diagnosis of thermal system under different loading conditions
Qian et al. A diagnosis method for diesel engine wear fault based on grey rough set and SOM neural network
CN114970709B (en) Improved GA-based data-driven AHU multi-fault diagnosis feature selection method
Wang et al. Research on the steam power system operation condition assessment method based on PCA-AE
Ma et al. Fuzzy neural network approach for fault diagnosis of power plant thermal system under different operating points
Kaiadi Artificial neural networks modelling for monitoring and performance analysis of a heat and power plant
CN117668460A (en) Superheated steam temperature deep learning modeling feature screening method based on SHAP evaluation
Alketbi et al. A Systems Engineering Approach to Predict the Success Window of FLEX Strategy under Extended SBO Using Artificial Intelligence
Arriagada et al. Artificial neural network model for a biomass-fueled boiler

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