CN111444659A - Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering - Google Patents

Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering Download PDF

Info

Publication number
CN111444659A
CN111444659A CN202010221861.8A CN202010221861A CN111444659A CN 111444659 A CN111444659 A CN 111444659A CN 202010221861 A CN202010221861 A CN 202010221861A CN 111444659 A CN111444659 A CN 111444659A
Authority
CN
China
Prior art keywords
particle swarm
particle
sample distribution
target
variation
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
CN202010221861.8A
Other languages
Chinese (zh)
Inventor
陈汉新
王琪
柯耀
苗育茁
黄浪
方璐
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.)
Wuhan Institute of Technology
Original Assignee
Wuhan Institute of Technology
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 Wuhan Institute of Technology filed Critical Wuhan Institute of Technology
Priority to CN202010221861.8A priority Critical patent/CN111444659A/en
Publication of CN111444659A publication Critical patent/CN111444659A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—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
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Structures Of Non-Positive Displacement Pumps (AREA)

Abstract

The invention relates to a centrifugal pump fault diagnosis method, a system and a medium based on improved particle filtering, which comprises the steps of establishing a standard comparison impeller model, and acquiring an original fault signal of a centrifugal pump impeller to be detected by using the standard comparison impeller model; filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal; performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal; and constructing a BP neural network identification model, and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result. The method effectively avoids the problems of particle scarcity and easy falling into local extreme values of the traditional particle filter algorithm, improves the convergence speed and precision, and effectively improves the efficiency and accuracy of fault diagnosis of the centrifugal pump impeller.

Description

Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering
Technical Field
The invention relates to the technical field of centrifugal pump fault diagnosis, in particular to a centrifugal pump fault diagnosis method, system and medium based on improved particle filtering.
Background
With the rapid development of scientific technology, the equipment fault diagnosis technology is gradually mature, and is developed vigorously in the field of comprehensive engineering as a new marginal subject. As a basic measure for ensuring the safe operation of the equipment, the fault diagnosis can effectively forecast the early fault development level of the equipment, judge the cause of fault formation, analyze the cause and propose a countermeasure proposal to treat the existing hidden danger so as to avoid or reduce the occurrence of accidents.
Centrifugal pumps are widely used general-purpose machines, and pumps are being developed in the direction of heavy-duty, high-speed, and light-duty pumps. The working strength of the pump is continuously improved, the working conditions are more and more severe, and the incidence rate of various faults of the pump is continuously increased. Accidents caused by pump failure are often catastrophic and cause significant losses. Therefore, it is important to ensure safe and reliable operation.
In a rotary machine, equipment state information is hidden in a rotor vibration signal and includes information on various abnormalities or faults of the equipment, and vibration parameters are important indexes for extracting fault characteristics. The centrifugal pump has a severe working environment, and abnormal vibration and noise are often accompanied when an impeller fails. In order to accurately identify the fault type through the vibration signal, effective denoising processing needs to be performed on the signal so as to eliminate interference of noise on fault characteristics. The common denoising methods include wavelet analysis, empirical mode decomposition, independent component analysis and the like. These algorithms all face the problem of complex parameter optimization selection in industrial applications. The particle filter algorithm is an effective algorithm for processing nonlinear signals. The algorithm is widely applied to the fields of fault diagnosis, navigation positioning, wireless communication, robot positioning, visual tracking and the like, however, the traditional particle filtering algorithm is easy to fall into a local extreme value, and the convergence speed and precision are limited, so that the efficiency of fault diagnosis of the centrifugal pump impeller is low, and the accuracy is not enough.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a centrifugal pump fault method, system and medium based on improved particle filtering, which can effectively avoid the problems of particle scarcity and easy local extremum sinking of the traditional particle filtering algorithm, improve the convergence speed and precision, and effectively improve the efficiency and accuracy of fault diagnosis of the centrifugal pump impeller.
The technical scheme for solving the technical problems is as follows:
the centrifugal pump fault diagnosis method based on the improved particle filtering comprises the following steps:
step 1: establishing a standard comparison impeller model, and acquiring an original fault signal of a centrifugal pump impeller to be detected by using the standard comparison impeller model;
step 2: filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
and step 3: performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
and 4, step 4: constructing a BP neural network identification model, and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
wherein, the step 2 specifically comprises:
step 21: taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
step 22: optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
step 23: obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
step 24: and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
According to another aspect of the present invention, there is also provided a centrifugal pump fault diagnosis system based on improved particle filtering, including a signal acquisition module, a filtering module, a dimensionality reduction module and an identification module;
the signal acquisition module is used for establishing a standard comparison impeller model and acquiring an original fault signal of the centrifugal pump impeller to be detected by using the standard comparison impeller model;
the filtering module is used for filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
the dimensionality reduction module is used for performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
the identification module is used for constructing a BP neural network identification model and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
the filtering module is specifically configured to:
taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
According to another aspect of the present invention, there is provided a centrifugal pump fault diagnosis system based on improved particle filtering, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the computer program is executed to implement the steps of a centrifugal pump fault diagnosis method based on improved particle filtering according to the present invention.
In accordance with another aspect of the present invention, there is provided a computer storage medium comprising: at least one instruction which, when executed, implements the steps in an improved particle filter based centrifugal pump fault diagnosis method of the present invention.
The centrifugal pump fault diagnosis method, system and medium based on improved particle filtering have the beneficial effects that: by establishing a standard comparison impeller model, the failed impeller is conveniently compared with the standard comparison impeller model to obtain an original fault signal; an improved particle filtering method is introduced in the denoising and filtering process, initial particle swarm sample distribution is optimized, and the optimized particle swarm sample distribution is resampled by utilizing an importance weight set, so that the problems of particle scarcity and easy falling into local extreme values of the traditional particle filtering algorithm and the problem of single sample at the later sampling period are effectively avoided, the sampling process in the traditional particle filtering algorithm is optimized, the convergence speed and precision are improved, the improved particle filtering method is utilized to denoise and filter an original fault signal, the filtering precision and efficiency are improved, the obtained target fault signal is convenient for subsequent analysis, and further the fault identification precision and efficiency are improved; the time domain characteristic values are extracted and the dimension is reduced by using the principal component analysis method, the number of characteristic variables is greatly reduced, the burden of subsequent training and testing of the BP neural network is reduced, the fault identification efficiency is further improved, and finally, the faults are identified by using the BP neural network, so that the identification effect of the whole centrifugal pump impeller faults is effectively improved.
Drawings
FIG. 1 is a schematic flow chart of a method for diagnosing a fault of a centrifugal pump based on improved particle filtering according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating a process of obtaining a target fault signal according to a first embodiment of the present invention;
FIG. 3 is a schematic size diagram of a mud pump used in one embodiment of the present invention;
FIG. 4 is a diagram of the original signals of a normal impeller and the original fault signals of three faulty impellers according to a first embodiment of the present invention;
FIG. 5 is a graph of noise reduction signals of four impellers obtained by NPSO-PF filtering at 2000rmp in accordance with one embodiment of the present invention;
FIG. 6 is a graph of noise reduction signals of four impellers obtained by PSO-PF filtering at 2000rmp in the first embodiment of the present invention;
FIG. 7 is a graph of noise reduction signals of four impellers obtained by PF filtering at 2000rmp in the first embodiment of the present invention;
FIG. 8 is a schematic view of a complete flow chart of filtering by using an improved particle filtering method according to one embodiment of the present invention;
FIG. 9 is a schematic flow chart illustrating a process of obtaining a target detection signal according to a first embodiment of the present invention;
FIG. 10 is a diagram illustrating the calculated principal component contribution rate of a failed impeller according to an embodiment of the present invention;
fig. 11 is a schematic flow chart illustrating a fault identification result obtained in the first embodiment of the present invention;
FIGS. 12-1 to 12-4 are schematic diagrams illustrating four verification results of NPSO-PF filtered four-fold cross validation according to one embodiment of the present invention;
fig. 13 is a schematic structural diagram of a centrifugal pump fault diagnosis system based on improved particle filtering in the second embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
The present invention will be described with reference to the accompanying drawings.
Embodiment one, as shown in fig. 1 and fig. 2, a method for diagnosing a fault of a centrifugal pump based on improved particle filtering includes the following steps:
s1: establishing a standard comparison impeller model, and acquiring an original fault signal of a centrifugal pump impeller to be detected by using the standard comparison impeller model;
s2: filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
s3: performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
s4: constructing a BP neural network identification model, and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
wherein, S2 specifically includes:
s21: taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
s22: optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
s23: obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
s24: and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
By establishing a standard comparison impeller model, the failed impeller is conveniently compared with the standard comparison impeller model to obtain an original fault signal; an improved particle filtering method is introduced in the denoising and filtering process, initial particle swarm sample distribution is optimized, and the optimized particle swarm sample distribution is resampled by utilizing an importance weight set, so that the problems of particle scarcity and easy falling into local extreme values of the traditional particle filtering algorithm and the problem of single sample at the later sampling period are effectively avoided, the sampling process in the traditional particle filtering algorithm is optimized, the convergence speed and precision are improved, the improved particle filtering method is utilized to denoise and filter an original fault signal, the filtering precision and efficiency are improved, the obtained target fault signal is convenient for subsequent analysis, and further the fault identification precision and efficiency are improved; the time domain characteristic values are extracted and the dimension is reduced by using the principal component analysis method, the number of characteristic variables is greatly reduced, the burden of subsequent training and testing of the BP neural network is reduced, the fault identification efficiency is further improved, and finally, the faults are identified by using the BP neural network, so that the identification effect of the whole centrifugal pump impeller faults is effectively improved.
Specifically, the embodiment uses a Weir/Warman 3/2CAH type slurry pump, the type of the matched closed impeller is C2147, the diameter of the impeller is 8.5 inches, the impeller is provided with 5 blades, and the size parameters of the slurry pump are shown in FIG. 3 and are in mm.
A normal impeller S1 and three impellers S2, S3 and S4 with damage are selected for experiments, a data model obtained by detecting the normal impeller S1 is a standard control impeller model, and the representation methods of the four impellers are specifically shown in Table 1.
TABLE 1 impeller presentation method
Figure BDA0002426370640000071
In this embodiment, the data acquisition system adopts an SCXI signal conditioning system, and the system includes 12 channels of signals, such as vibration signals, sound signals, temperature, pressure, and the like; a Dell Inspiron9200 notebook computer is used for storing data; the sampling rate of the system is 9kHZ, and the sampling time is 20 s; in addition, 3 triaxial vibration acceleration sensors are respectively arranged at the top of the pump, the outlet of the pump and the top of the bearing; the working medium is clear water, and experimental data of four impellers in a stable operation state at 2000rpm are recorded respectively.
Preferably, S21 specifically includes:
s211: taking the original fault signal as the particle swarm, and setting initialization parameters of the particle swarm;
the initialization parameters comprise the number of particle swarms, learning factors, maximum iteration times, control coefficients, maximum inertia coefficients, minimum inertia coefficients, preset variation rates, and initial positions and initial speeds of particles;
s212: initializing the particle swarm according to the initialization parameters to obtain the initial particle swarm sample distribution;
the initial particle swarm sample distribution is:
Figure BDA0002426370640000081
wherein the content of the first and second substances,
Figure BDA0002426370640000082
for the initial particle population sample distribution,
Figure BDA0002426370640000083
is the ith sample point in the initial particle swarm sample distribution, and N is the particle swarm number.
By setting the initialization parameters of the particle swarm, initial particle swarm sample distribution can be obtained conveniently according to the initialization parameters, and subsequent optimization and variation operation can be carried out according to the initialization parameters, so that the filtering efficiency and precision can be improved conveniently.
Specifically, the data acquisition time of each impeller in the present embodiment is 20s, and a total of 180000 data points. The raw signal plot for a normal impeller and the raw fault signal plots for three failed impellers are shown in fig. 4.
The original signal equation is:
Figure BDA0002426370640000084
the random noise signal is added as follows:
Z(t)＝Y(t)+0.005R(t)；
the initialization parameter portion of the settings is shown in table 2.
Initialization parameters set in Table 2
Figure BDA0002426370640000085
Figure BDA0002426370640000091
Preferably, S22 specifically includes:
s221: calculating to obtain a variation control function according to the maximum iteration times and the control coefficient;
the variation control function is:
Figure BDA0002426370640000092
wherein y (epoch) is the variation control function, epoch is the current iteration number, epochmaxα and β are both the control coefficients for the maximum number of iterations;
s222: calculating to obtain a dynamic inertia coefficient according to the control variation function, the maximum inertia coefficient and the minimum inertia coefficient;
the formula for calculating the dynamic inertia coefficient is specifically as follows:
w＝wmin+(wmax-wmin)·y(epoch)；
wherein w is the dynamic inertia coefficient, wminIs the minimum coefficient of inertia, wmaxIs the maximum coefficient of inertia;
s223: calculating to obtain variation rate according to the variation control function and the preset variation rate, calculating to obtain the number of variation particles according to the variation rate, and performing variation operation on the particle swarm in the initial particle swarm sample distribution according to the number of variation particles and the variation control function to obtain a variation particle swarm;
the formula for calculating the variation rate is specifically as follows:
u＝m·y(epoch)；
the formula for calculating the number of the variant particles is specifically as follows:
M＝N·u；
wherein u is the variation rate, M is the preset variation rate, and M is the number of the variation particles;
the formula for performing mutation operation on the ith particle in the particle swarm is specifically as follows:
X′ij＝Xij+r1·y(epoch)；
wherein, XijIs the position, X 'of the jth element of the ith particle in the particle swarm'ijThe position r of a variation particle corresponding to the j element of the ith particle in the particle swarm after variation operation1Is a random number;
s224: obtaining the fitness value of the variation particle swarm, and obtaining the individual optimal value and the global optimal value of the variation particle swarm in each iteration according to the fitness value;
the formula for calculating the fitness value is specifically as follows:
Figure BDA0002426370640000101
wherein fit is the fitness value, RtTo observe the variance of the noise, YnewAs a current observed value, YpredTo predict an observed value;
s225: updating the variation particle swarm according to the individual optimal value, the global optimal value, the dynamic inertia coefficient, the learning factor and the maximum iteration number to obtain the optimized particle swarm sample distribution;
the formula for updating the group of variation particles comprises:
Figure BDA0002426370640000102
wherein the content of the first and second substances,
Figure BDA0002426370640000103
the speed of the ith particle in the variation particle swarm after the (k + 1) th iteration updating,
Figure BDA0002426370640000104
the updated speed of the ith particle in the variation particle swarm in the k iteration is obtained,
Figure BDA0002426370640000105
for the individual optima for the k-th iteration,
Figure BDA0002426370640000106
for the global optimum of the k-th iteration,
Figure BDA0002426370640000107
for the position of the ith particle in the variation particle swarm after the (k + 1) th iteration updating,
Figure BDA0002426370640000108
is the position of the ith particle in the variation particle swarm after the k iteration update, r2Is a random number between 0 and 1, c1And c2Are all the learning factors.
The traditional particle filter method (PF algorithm) is based on the optimal Bayesian estimation of a Monte Carlo method, and the particle swarm optimization method (PSO algorithm) is a random search method for simulating foraging behavior of a bird swarm. The PSO algorithm with the mutation operator is integrated into the PF algorithm, and the sampling process of the PF is optimized by the fast and accurate convergence performance of the PSO algorithm with the mutation operator; all the particles are moved to the optimal particles through the fitness value, so that the distribution condition of the particle swarm is improved, the particle swarm is concentrated near the real state, and the precision and the efficiency of filtering are improved.
Preferably, S23 specifically includes:
s231: obtaining an importance density function according to the optimized particle swarm sample distribution, calculating an importance weight corresponding to each particle in the optimized particle swarm sample distribution one by one according to the importance density function, and obtaining an importance weight set according to the importance weights of all the particles;
the formula for calculating the importance weight of the ith particle in the kth iteration is specifically as follows:
Figure BDA0002426370640000111
wherein the content of the first and second substances,
Figure BDA0002426370640000112
for the importance weight of the ith particle in the optimized particle swarm sample distribution at the kth iteration,
Figure BDA0002426370640000113
for the dynamic inertia coefficient of the ith particle in the optimized particle swarm sample distribution at the k-1 iteration,
Figure BDA0002426370640000114
for the state quantity of the ith particle in the optimized particle swarm sample distribution at the kth iteration,
Figure BDA0002426370640000115
for the state quantity of the ith particle in the optimized particle swarm sample distribution at the k-1 th iteration,
Figure BDA0002426370640000116
for an observed quantity of an ith particle in the optimized particle swarm sample distribution at a kth iteration,
Figure BDA0002426370640000117
for the first a posteriori probability density function,
Figure BDA0002426370640000118
is a function of the second a posteriori probability density,
Figure BDA0002426370640000119
is the importance density function;
s232: normalizing each importance weight in the importance weight set respectively to obtain a normalized weight corresponding to each importance weight one by one;
s233: resampling the optimized particle swarm sample distribution according to all normalized importance weights to obtain the target particle swarm sample distribution;
s234: and obtaining target importance weights corresponding to the normalized importance weights one by one according to a preset importance weight, and obtaining the target importance weight set corresponding to the target particle swarm sample distribution according to all the target importance weights.
Through calculation and normalization of the importance weight set, subsequent resampling for optimizing the distribution of the particle swarm samples is facilitated, and the distribution condition of the particle swarm is further improved; when the importance density function replaces the posterior probability distribution to be used as a sampling function, the variance of the importance weight is increased along with the time, so that the importance weights of the particles are gathered on a few particles, after several recursions, the importance weights of the particles are distributed in an unbalanced manner, and finally only one particle possibly has a nonzero weight, thereby causing the problem that the updated particle set cannot correctly reflect the actual situation of the posterior probability distribution, namely the particle degradation; according to the invention, the new particle importance weight (namely a target importance weight set which comprises a plurality of target importance weights) after resampling is calculated by presetting the importance weights, so that particle degradation is effectively solved, and the filtering and denoising effects are improved by an improved particle filtering method.
Specifically, in this embodiment, in order to compare and verify the filtering effect of the improved particle filtering method (i.e., NPSO-PF algorithm), two other filtering methods are further adopted to respectively filter the original signal of the normal impeller and the original fault signal of the three faulty impellers at the rotational speed of the centrifugal pump at 2000rpm, specifically, the PSO-PF filtering method and the PF filtering method, where the noise reduction signal graphs of the four impellers obtained by adopting the NPSO-PF filtering method at 2000rmp rotational speed are shown in fig. 5, the noise reduction signal graphs of the four impellers obtained by adopting the PSO-PF filtering method at 2000rmp rotational speed are shown in fig. 6, and the noise reduction signal graphs of the four impellers obtained by adopting the PF filtering method at 2000rmp rotational speed are shown in fig. 7.
The PF filtering method is used for realizing recursive Bayesian filtering by a non-parametric Monte Carlo (Monte Carlo) simulation method, is suitable for any nonlinear system which can be described by a state space model, and can approximate to optimal estimation in precision; the PSO-PF filtering method is characterized in that a PSO algorithm with a mutation operator is integrated into a PF algorithm, and the sampling process of the PF is optimized by the fast and accurate convergence performance of the PSO algorithm with the mutation operator. All the particles are moved to the optimal particles through the fitness value so as to improve the distribution condition of the particle sets, and the particle sets are concentrated near the real state, so that the precision and the efficiency of filtering are improved; the NPSO-PF filtering method (i.e. the improved particle filtering method adopted in the invention) is an algorithm which optimizes a particle updating mode, introduces an advantage speed and a disadvantage speed, improves a local optimal phenomenon and improves an optimized solution on the basis of a PSO-PF algorithm.
As can be seen from fig. 4 to 7, the NPSO-PF filtering method in the present embodiment has the best filtering effect.
Specifically, the complete flow of filtering with the improved particle filtering method in this embodiment is shown in fig. 8.
Preferably, as shown in fig. 9, S3 includes:
s31: performing time domain analysis on the target fault signal by using a principal component analysis method, and extracting a plurality of time domain characteristic variables;
s32: making all time domain characteristic variables into a characteristic variable sample set;
s33: calculating a sample mean value and a sample variable covariance matrix of the characteristic variable sample set, and obtaining characteristic values corresponding to each time domain characteristic variable one by one according to the sample mean value and the sample variable covariance matrix;
s34: respectively calculating the principal component contribution rate and the accumulated contribution rate corresponding to each principal component according to the characteristic values of all the time domain characteristic variables by using the principal component analysis method;
the formula for calculating the principal component contribution rate of the jth principal component is specifically as follows:
Figure BDA0002426370640000131
the formula for calculating the cumulative contribution rate of the jth principal element is specifically as follows:
Figure BDA0002426370640000132
wherein, ηjPrincipal component contribution, λ, for the jth principal componentjIs the characteristic value, lambda, corresponding to the jth time domain characteristic variabletIs the characteristic value corresponding to the t time domain characteristic variable, n is the time domain characteristic variable number, βjThe cumulative contribution rate for the jth principal element;
s35: and obtaining a principal element mapping matrix according to the accumulated contribution rate of all principal elements, and transforming all time domain characteristic variables by using the principal element mapping matrix to obtain the target detection signal.
The Principal Component Analysis (PCA) is based on an original data space, the dimension of the original data space is reduced by constructing a group of new latent variables, main change information is extracted from a new mapping space, and statistical characteristics are extracted, so that the understanding of the space characteristics of the original data is formed; according to the invention, through the extraction of the time domain feature vector and the calculation of the corresponding feature value, the principal component contribution rate and the accumulated contribution rate of each principal component can be conveniently calculated subsequently, and through the analysis of all the accumulated contribution rates, the principal component mapping matrix can be conveniently obtained, so that the feature space of the time domain feature variable can be obtained, and further the time domain feature vector can be transformed, and the dimension reduction of the target fault signal can be realized.
It should be noted that, in the present embodiment, a specific formula for calculating the sample mean and the sample vector covariance matrix is the prior art, and is not described herein again.
Specifically, the time domain feature variables extracted in this embodiment include: the average value, effective value, standard deviation, maximum value, minimum value, peak value, skewness coefficient, peak value factor, kurtosis factor, margin factor, waveform index, kurtosis index and pulse index are 13 time domain statistical indexes. The noise reduction signal diagram of the normal impeller after noise reduction and the target fault signal of the three fault impellers after noise reduction both comprise 180000 data points, 4500 data points are used as one group and are divided into 40 groups for analysis, and the total number of the groups is 160, so that the noise reduction signal diagram of the normal impeller after noise reduction and the target fault signal of the three fault impellers after noise reduction both comprise 13 time domain characteristic variables, and each time domain characteristic variable comprises 160-dimensional data.
The 13 time-domain characteristic variables of one of the failed impellers are calculated according to the steps from S31 to S34, and the obtained principal component contribution rate and the accumulated contribution rate of each principal component are shown in table 3 and fig. 10.
TABLE 3 principal component contribution rate and cumulative contribution rate calculated for target fault signal after NPSO-PF noise reduction
Figure BDA0002426370640000141
Figure BDA0002426370640000151
It can be seen from table 3 and fig. 10 that the maximum first principal component contribution rate of the signal after being subjected to the NPSO-PF noise reduction treatment is 99.38% by PCA analysis, and the cumulative contribution rates of the first five principal components are up to 100%, and the result shows that the NPSO-PF noise reduction effect is the best compared with the PSO-PF and PF, the influence on the subsequent PCA treatment is the greatest, and the number of characteristic variables is greatly reduced after the PCA treatment.
Preferably, as shown in fig. 11, S4 includes:
s41: acquiring a data set, and acquiring a training set from the data set;
s42: determining model parameters, and constructing a BP neural network training model according to the model parameters;
s43: training the BP neural network training model by using the training set to obtain the BP neural network identification model;
s44: carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
the model parameters comprise a transfer function, an input layer node number, an output layer node number and a hidden layer node number.
The BP neural network is a multilayer feedforward neural network trained according to an error reverse propagation algorithm, a BP neural network training model is established by determining model parameters such as a transfer function, the number of nodes of an input layer, the number of nodes of an output layer, the number of nodes of a hidden layer and the like, then training is carried out by utilizing a training set, and the obtained BP neural network identification model identifies a target detection signal obtained in S3, so that the efficiency is high, and the precision is high.
Specifically, the sigmoid transfer function in this embodiment is:
Figure BDA0002426370640000152
the number of nodes in the input layer is equal to the observed quantity of the input samples, and the first five principal elements after the PCA analysis comprise 100% of characteristic information, so the number l of the nodes in the input layer is selected1(ii) 5; because four impeller state modes are set, the number of nodes of the output layer is l 24; the number of hidden layer nodes has great influence on the performance of the BP network, the larger the number of hidden layer nodes is, the stronger the learning capacity of the BP network is, but the too large number of nodes can slow the convergence rate, the smaller the number of nodes is, the faster the convergence rate is, but the learning capacity of the BP network is very low, and the fault cannot be accurately identified, so the number l of hidden layer nodes is3The determination formula of (1) is as follows:
Figure BDA0002426370640000161
since a is [1,9 ]]Constant between, thus implying the number of layer nodes l3In the range of [4,12 ]]。
Specifically, in this embodiment, 160 groups of data of the four impellers are used as a data set, 40 groups of data are randomly extracted from the data set as a training set, and the training set is input into an established BP neural network training model for training, so that training errors obtained under different hidden layer node numbers are shown in table 4.
TABLE 4 number of nodes of different hidden layers l3Error in training
l3Value of 4 5 6 7 8 9 10 11 12
Error in training 0.050 0.008 0.043 0.009 0.015 0.104 0.037 0.034 0.033
As can be seen from Table 4, when the number of hidden layer nodes l3When the training error is minimum at 5, the number of hidden layer nodes of the BP neural network training model is determined to be 5.
Preferably, after S43, the method further comprises:
and verifying the BP neural network identification model by adopting a k-fold cross verification method, if the verification is passed, executing S44, and if the verification is failed, returning to S42.
The BP neural network identification model is verified through a k-fold cross verification method, and the effectiveness and the accuracy of the model in identifying the centrifugal pump faults can be effectively guaranteed.
Specifically, the k-fold cross validation method in this embodiment is a four-fold cross validation method, and the process is to divide the data set into four parts, one part is used for testing, the remaining three parts are used for training the BP neural network training model, and the validation accuracy is an average value of the validation accuracy rate of each time. In this embodiment, 160 sets of sample data after being denoised by three filtering methods and optimized by PCA are randomly divided into 4 parts, and the 4 parts are substituted into the constructed BP neural network training model to perform four-fold cross validation, wherein four validation results of four-fold cross validation after NPSO-PF filtering are shown in fig. 12-1 to 12-4, and comparison conditions of cross validation results after being denoised by the three filtering methods are shown in table 5.
TABLE 5 comparison of cross-validation results after denoising by three filtering methods
Figure BDA0002426370640000171
As can be seen from Table 5, after the denoising and PCA analysis by the three filtering methods, the identification accuracy of the centrifugal pump impeller fault identification exceeds 90%, wherein the identification accuracy of the NPSO-PF-BP mode (i.e. the method in the invention) on the centrifugal pump impeller fault identification is up to 99.375%, which is remarkably improved compared with 94.375% of the traditional particle filtering method.
In a second embodiment, as shown in fig. 13, the centrifugal pump fault diagnosis system based on improved particle filtering includes a signal obtaining module, a filtering module, a dimension reduction module, and an identification module;
the signal acquisition module is used for establishing a standard comparison impeller model and acquiring an original fault signal of the centrifugal pump impeller to be detected by using the standard comparison impeller model;
the filtering module is used for filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
the dimensionality reduction module is used for performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
the identification module is used for constructing a BP neural network identification model and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
the filtering module is specifically configured to:
taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
The system for identifying the fault of the centrifugal pump of the embodiment is convenient for comparing the faulted impeller with the standard comparison impeller model by establishing the standard comparison impeller model, and obtains an original fault signal; an improved particle filtering method is introduced in the denoising and filtering process, initial particle swarm sample distribution is optimized, and the optimized particle swarm sample distribution is resampled by utilizing an importance weight set, so that the problems of particle scarcity and easy falling into local extreme values of the traditional particle filtering algorithm and the problem of single sample at the later sampling period are effectively avoided, the sampling process in the traditional particle filtering algorithm is optimized, the convergence speed and precision are improved, the improved particle filtering method is utilized to denoise and filter an original fault signal, the filtering precision and efficiency are improved, the obtained target fault signal is convenient for subsequent analysis, and further the fault identification precision and efficiency are improved; the time domain characteristic values are extracted and the dimension is reduced by using the principal component analysis method, the number of characteristic variables is greatly reduced, the burden of subsequent training and testing of the BP neural network is reduced, the fault identification efficiency is further improved, and finally, the faults are identified by using the BP neural network, so that the identification effect of the whole centrifugal pump impeller faults is effectively improved.
In a third embodiment, based on the first embodiment and the second embodiment, the present embodiment further discloses a centrifugal pump fault diagnosis system based on improved particle filtering, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the computer program is executed, the specific steps of S1 to S4 shown in fig. 1 are implemented.
The method realizes the fault recognition of the centrifugal pump by the computer program stored in the memory and running on the processor, effectively avoids the problems of particle scarcity and easy local extremum sinking of the traditional particle filter algorithm and the problem of single sample at the later sampling period based on the improved particle filter method, optimizes the sampling process in the traditional particle filter algorithm, improves the convergence speed and precision, extracts and reduces the dimension of a time domain characteristic value by using a principal component analysis method, reduces the burden of training and testing of a subsequent BP neural network, further improves the fault recognition efficiency, finally recognizes the fault by using the BP neural network, and effectively improves the recognition effect of the impeller fault of the whole centrifugal pump.
The present embodiment also provides a computer storage medium having at least one instruction stored thereon, where the instruction when executed implements the specific steps of S1-S4.
The method has the advantages that the problem that particles are poor and easy to fall into local extreme values in the traditional particle filter algorithm and the problem that samples in the later period of sampling are single are effectively solved based on the improved particle filter method, the sampling process in the traditional particle filter algorithm is optimized, the convergence speed and precision are improved, the principal component analysis method is used for extracting and reducing the dimension of time domain characteristic values, the burden of subsequent training and testing of a BP neural network is reduced, the fault identification efficiency is further improved, finally, the faults are identified by the BP neural network, and the identification effect of the impeller faults of the whole centrifugal pump is effectively improved.
Details of S1 to S4 in this embodiment are not described in detail in the first embodiment and the detailed descriptions in fig. 1 to fig. 12-4, which are not repeated herein.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. The centrifugal pump fault diagnosis method based on the improved particle filtering is characterized by comprising the following steps of:
step 1: establishing a standard comparison impeller model, and acquiring an original fault signal of a centrifugal pump impeller to be detected by using the standard comparison impeller model;
step 2: filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
and step 3: performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
and 4, step 4: constructing a BP neural network identification model, and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
wherein, the step 2 specifically comprises:
step 21: taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
step 22: optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
step 23: obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
step 24: and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
2. The centrifugal pump fault diagnosis method based on improved particle filtering as claimed in claim 1, wherein said step 21 specifically comprises:
step 211: taking the original fault signal as the particle swarm, and setting initialization parameters of the particle swarm;
the initialization parameters comprise the number of particle swarms, learning factors, maximum iteration times, control coefficients, maximum inertia coefficients, minimum inertia coefficients, preset variation rates, and initial positions and initial speeds of particles;
step 212: initializing the particle swarm according to the initialization parameters to obtain the initial particle swarm sample distribution;
the initial particle swarm sample distribution is:
Figure FDA0002426370630000021
wherein the content of the first and second substances,
Figure FDA0002426370630000022
for the initial particle population sample distribution,
Figure FDA0002426370630000023
is the ith sample point in the initial particle swarm sample distribution, and N is the particle swarm number.
3. A centrifugal pump fault diagnosis method based on improved particle filtering according to claim 2, characterized in that said step 22 specifically comprises:
step 221: calculating to obtain a variation control function according to the maximum iteration times and the control coefficient;
the variation control function is:
Figure FDA0002426370630000024
wherein y (epoch) is the variation control function, epoch is the current iteration number, epochmaxα and β are both the control coefficients for the maximum number of iterations;
step 222: calculating to obtain a dynamic inertia coefficient according to the control variation function, the maximum inertia coefficient and the minimum inertia coefficient;
the formula for calculating the dynamic inertia coefficient is specifically as follows:
w＝wmin+(wmax-wmin)·y(epoch)；
wherein w is the dynamic inertia coefficient, wminIs the minimum coefficient of inertia, wmaxIs the maximum coefficient of inertia;
step 223: calculating to obtain variation rate according to the variation control function and the preset variation rate, calculating to obtain the number of variation particles according to the variation rate, and performing variation operation on the particle swarm in the initial particle swarm sample distribution according to the number of variation particles and the variation control function to obtain a variation particle swarm;
the formula for calculating the variation rate is specifically as follows:
u＝m·y(epoch)；
the formula for calculating the number of the variant particles is specifically as follows:
M＝N·u；
wherein u is the variation rate, M is the preset variation rate, and M is the number of the variation particles;
the formula for performing mutation operation on the ith particle in the particle swarm is specifically as follows:
X′ij＝Xij+r1·y(epoch)；
wherein, XijIs the position of the jth element of the ith particle in the particle population, Xi′jThe position r of a variation particle corresponding to the j element of the ith particle in the particle swarm after variation operation1Is a random number;
step 224: obtaining the fitness value of the variation particle swarm, and obtaining the individual optimal value and the global optimal value of the variation particle swarm in each iteration according to the fitness value;
the formula for calculating the fitness value is specifically as follows:
Figure FDA0002426370630000031
wherein fit is the fitness value, RtTo observe the variance of the noise, YnewAs a current observed value, YpredTo predict an observed value;
step 225: updating the variation particle swarm according to the individual optimal value, the global optimal value, the dynamic inertia coefficient, the learning factor and the maximum iteration number to obtain the optimized particle swarm sample distribution;
the formula for updating the group of variation particles comprises:
Figure FDA0002426370630000041
wherein, Vi′k+1The velocity, V, of the ith particle in the variation particle swarm after the (k + 1) th iteration updatei′kThe velocity, P, of the ith particle in the variation particle swarm after the k iteration updatei′kFor the individual optima for the k-th iteration,
Figure FDA0002426370630000042
for the global optimum of the k-th iteration,
Figure FDA0002426370630000043
for the position of the ith particle in the variation particle swarm after the (k + 1) th iteration updating,
Figure FDA0002426370630000044
is the position of the ith particle in the variation particle swarm after the k iteration update, r2Is a random number between 0 and 1, c1And c2Are all the learning factors.
4. A centrifugal pump fault diagnosis method based on improved particle filtering according to claim 3, characterized in that said step 23 specifically comprises:
step 231: obtaining an importance density function according to the optimized particle swarm sample distribution, calculating an importance weight corresponding to each particle in the optimized particle swarm sample distribution one by one according to the importance density function, and obtaining an importance weight set according to the importance weights of all the particles;
the formula for calculating the importance weight of the ith particle in the kth iteration is specifically as follows:
Figure FDA0002426370630000045
wherein the content of the first and second substances,
Figure FDA0002426370630000046
for the importance weight of the ith particle in the optimized particle swarm sample distribution at the kth iteration,
Figure FDA0002426370630000047
for the dynamic inertia coefficient of the ith particle in the optimized particle swarm sample distribution at the k-1 iteration,
Figure FDA0002426370630000048
for the state quantity of the ith particle in the optimized particle swarm sample distribution at the kth iteration,
Figure FDA0002426370630000049
for the state quantity of the ith particle in the optimized particle swarm sample distribution at the k-1 th iteration,
Figure FDA00024263706300000410
for an observed quantity of an ith particle in the optimized particle swarm sample distribution at a kth iteration,
Figure FDA00024263706300000411
for the first a posteriori probability density function,
Figure FDA00024263706300000412
is a function of the second a posteriori probability density,
Figure FDA00024263706300000413
is the importance density function;
step 232: normalizing each importance weight in the importance weight set respectively to obtain a normalized weight corresponding to each importance weight one by one;
step 233: resampling the optimized particle swarm sample distribution according to all normalized importance weights to obtain the target particle swarm sample distribution;
step 234: and obtaining target importance weights corresponding to the normalized importance weights one by one according to a preset importance weight, and obtaining the target importance weight set corresponding to the target particle swarm sample distribution according to all the target importance weights.
5. The improved particle filter based centrifugal pump fault diagnosis method of claim 1, wherein said step 3 comprises:
step 31: performing time domain analysis on the target fault signal by using a principal component analysis method, and extracting a plurality of time domain characteristic variables;
step 32: making all time domain characteristic variables into a characteristic variable sample set;
step 33: calculating a sample mean value and a sample variable covariance matrix of the characteristic variable sample set, and obtaining characteristic values corresponding to each time domain characteristic variable one by one according to the sample mean value and the sample variable covariance matrix;
step 34: respectively calculating the principal component contribution rate and the accumulated contribution rate corresponding to each principal component according to the characteristic values of all the time domain characteristic variables by using the principal component analysis method;
the formula for calculating the principal component contribution rate of the jth principal component is specifically as follows:
Figure FDA0002426370630000051
the formula for calculating the cumulative contribution rate of the jth principal element is specifically as follows:
Figure FDA0002426370630000052
wherein, ηjPrincipal component contribution, λ, for the jth principal componentjIs the jthCharacteristic value, lambda, corresponding to the time-domain characteristic variabletIs the characteristic value corresponding to the t time domain characteristic variable, n is the time domain characteristic variable number, βjThe cumulative contribution rate for the jth principal element;
step 35: and obtaining a principal element mapping matrix according to the accumulated contribution rate of all principal elements, and transforming all time domain characteristic variables by using the principal element mapping matrix to obtain the target detection signal.
6. The improved particle filter based centrifugal pump fault diagnosis method of claim 1, wherein said step 4 comprises:
step 41: acquiring a data set, and acquiring a training set from the data set;
step 42: determining model parameters, and constructing a BP neural network training model according to the model parameters;
step 43: training the BP neural network training model by using the training set to obtain the BP neural network identification model;
step 44: carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
the model parameters comprise a transfer function, an input layer node number, an output layer node number and a hidden layer node number.
7. A centrifugal pump fault diagnosis method based on improved particle filtering according to claim 6, further comprising, after said step 43:
and verifying the BP neural network identification model by adopting a k-fold cross verification method, if the verification is passed, executing the step 44, and if the verification is failed, returning to the step 42.
8. The centrifugal pump fault diagnosis system based on improved particle filtering is characterized by comprising a signal acquisition module, a filtering module, a dimension reduction module and an identification module;
the signal acquisition module is used for establishing a standard comparison impeller model and acquiring an original fault signal of the centrifugal pump impeller to be detected by using the standard comparison impeller model;
the filtering module is used for filtering the original fault signal by using an improved particle filtering method to obtain a target fault signal;
the dimensionality reduction module is used for performing time domain analysis on the target fault signal by using a principal component analysis method, extracting a plurality of time domain characteristic variables, and performing dimensionality reduction processing on all the time domain characteristic variables by using the principal component analysis method to obtain a target detection signal;
the identification module is used for constructing a BP neural network identification model and carrying out fault identification on the target detection signal by using the BP neural network identification model to obtain a fault identification result;
the filtering module is specifically configured to:
taking the original fault signal as a particle swarm, initializing the particle swarm to obtain initial particle swarm sample distribution;
optimizing the initial particle swarm sample distribution to obtain optimized particle swarm sample distribution;
obtaining an importance weight set of the optimized particle swarm sample distribution, and resampling the optimized particle swarm sample distribution according to the importance weight set to obtain a target particle swarm sample distribution and a target importance weight set corresponding to the target particle swarm sample distribution;
and carrying out global search according to the target particle swarm sample distribution and the target importance weight set to obtain the target fault signal.
9. Centrifugal pump fault diagnosis system based on improved particle filtering, characterized in that it comprises a processor, a memory and a computer program stored in said memory and executable on said processor, said computer program realizing the method steps of any of claims 1 to 7 when executed.
10. A computer storage medium, the computer storage medium comprising: at least one instruction which, when executed, implements the method steps of any one of claims 1 to 7.
CN202010221861.8A 2020-03-26 2020-03-26 Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering Pending CN111444659A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010221861.8A CN111444659A (en) 2020-03-26 2020-03-26 Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010221861.8A CN111444659A (en) 2020-03-26 2020-03-26 Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering

Publications (1)

Publication Number Publication Date
CN111444659A true CN111444659A (en) 2020-07-24

Family

ID=71648985

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010221861.8A Pending CN111444659A (en) 2020-03-26 2020-03-26 Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering

Country Status (1)

Country Link
CN (1) CN111444659A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113449409A (en) * 2020-09-03 2021-09-28 鲁能集团有限公司 Method and equipment for storing sample data of offshore wind turbine fault diagnosis model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104361238A (en) * 2014-11-17 2015-02-18 北京信息科技大学 Fault sensitive characteristic extraction method based on information entropy improved PCA (Principal Component Analysis)
CN110146279A (en) * 2019-05-21 2019-08-20 杭州电子科技大学 A kind of marine shafting imbalance fault diagnostic method based on vector evidential reasoning
CN110263907A (en) * 2019-07-01 2019-09-20 上海海事大学 Based on the ship short trouble diagnostic method for improving GA-PSO-BP
CN110390272A (en) * 2019-06-30 2019-10-29 天津大学 A kind of EEG signal feature dimension reduction method based on weighted principal component analyzing

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104361238A (en) * 2014-11-17 2015-02-18 北京信息科技大学 Fault sensitive characteristic extraction method based on information entropy improved PCA (Principal Component Analysis)
CN110146279A (en) * 2019-05-21 2019-08-20 杭州电子科技大学 A kind of marine shafting imbalance fault diagnostic method based on vector evidential reasoning
CN110390272A (en) * 2019-06-30 2019-10-29 天津大学 A kind of EEG signal feature dimension reduction method based on weighted principal component analyzing
CN110263907A (en) * 2019-07-01 2019-09-20 上海海事大学 Based on the ship short trouble diagnostic method for improving GA-PSO-BP

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HANXIN CHEN.ET AL.: "Centrifugal Pump Impeller Health Diagnosis Based on Improved Particle Filter and BP Neural Network", pages 217 - 224 *
方璐等: "一种带变异算子的粒子群优化粒子滤波降噪算法", vol. 41, no. 4, pages 392 - 398 *
白晓波: "非线性系统的状态估计方法", 西南交通大学出版社, pages: 97 - 100 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113449409A (en) * 2020-09-03 2021-09-28 鲁能集团有限公司 Method and equipment for storing sample data of offshore wind turbine fault diagnosis model
CN113449409B (en) * 2020-09-03 2022-10-04 中国绿发投资集团有限公司 Method and equipment for storing sample data of offshore wind turbine fault diagnosis model

Similar Documents

Publication Publication Date Title
CN110849627B (en) Width migration learning network and rolling bearing fault diagnosis method based on same
CN106682688B (en) Particle swarm optimization-based stacked noise reduction self-coding network bearing fault diagnosis method
CN113923104B (en) Network fault diagnosis method, equipment and storage medium based on wavelet neural network
CN110348468B (en) Bearing sub-health identification method for optimized strong reconstruction edge noise reduction automatic encoder of correlation vector machine
CN114399032B (en) Method and system for predicting metering error of electric energy meter
CN111043050A (en) Fault diagnosis method and system for centrifugal pump
CN115017945A (en) Mechanical fault diagnosis method and system based on enhanced convolutional neural network
Wang et al. Fault diagnosis of centrifugal pump using symptom parameters in frequency domain
WO2022222026A1 (en) Medical diagnosis missing data completion method and completion apparatus, and electronic device and medium
CN114692758A (en) Power communication fault analysis method and device, terminal equipment and medium
CN115901249B (en) Rolling bearing performance degradation evaluation method combining feature optimization and multi-strategy optimization SVDD
CN112101142A (en) Slurry pump running state evaluation method, monitoring terminal and computer readable storage medium
CN111521398A (en) Gear box fault diagnosis method and system based on BP neural network and principal component analysis method
CN112504682A (en) Chassis engine fault diagnosis method and system based on particle swarm optimization algorithm
CN114091525A (en) Rolling bearing degradation trend prediction method
Ge et al. An improved PF remaining useful life prediction method based on quantum genetics and LSTM
CN111444659A (en) Centrifugal pump fault diagnosis method, system and medium based on improved particle filtering
CN114741945A (en) Diagnosis method for online faults of aircraft engine
CN117150882A (en) Engine oil consumption prediction method, system, electronic equipment and storage medium
CN116881841A (en) Hybrid model fault diagnosis method based on F1-score multistage decision analysis
CN113408182B (en) Tool life full-cycle wear diagnosis method, device and storage medium based on multiple wavelet optimal features and neural network
CN113048012B (en) Wind turbine generator yaw angle identification method and device based on Gaussian mixture model
CN113051809A (en) Virtual health factor construction method based on improved restricted Boltzmann machine
CN103197668B (en) Data drive type control performance detection device and method
CN113188235A (en) Multi-source fault diagnosis method, system and equipment for sensor in air conditioning system

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