CN109240276B - Multi-block PCA fault monitoring method based on fault sensitive principal component selection - Google Patents

Multi-block PCA fault monitoring method based on fault sensitive principal component selection Download PDF

Info

Publication number
CN109240276B
CN109240276B CN201811330777.9A CN201811330777A CN109240276B CN 109240276 B CN109240276 B CN 109240276B CN 201811330777 A CN201811330777 A CN 201811330777A CN 109240276 B CN109240276 B CN 109240276B
Authority
CN
China
Prior art keywords
fault
monitoring
block
sub
pca
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.)
Active
Application number
CN201811330777.9A
Other languages
Chinese (zh)
Other versions
CN109240276A (en
Inventor
熊伟丽
顾炳斌
马君霞
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.)
Jiangnan University
Original Assignee
Jiangnan 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 Jiangnan University filed Critical Jiangnan University
Priority to CN201811330777.9A priority Critical patent/CN109240276B/en
Publication of CN109240276A publication Critical patent/CN109240276A/en
Application granted granted Critical
Publication of CN109240276B publication Critical patent/CN109240276B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
    • G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
    • G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses a multi-block PCA fault monitoring method based on fault sensitive principal component selection, which aims at the problem of how to select principal components in the traditional PCA fault monitoring algorithm, defines a fault sensitive coefficient epsilon as a new principal component sorting criterion, has m sorting results for the principal components based on m-dimensional variables, and divides each sorting result into a sub-block. Selecting the fault sensitivity coefficient epsilon larger than the threshold epsilon in each sub-blocklimThe principal element of (1) carries out fault monitoring and calculates T of each sub-block2Statistics are obtained. And fusing the monitoring results of the sub-blocks by using a Bayesian inference method to obtain a final BIC monitoring result. According to the method, on one hand, principal elements can be extracted without the aid of a fault data set, and on the other hand, computing resources consumed by real-time modeling are avoided.

Description

Multi-block PCA fault monitoring method based on fault sensitive principal component selection
Technical Field
The invention relates to a fault sensitive principal component selection-based multi-block PCA fault monitoring method, and belongs to the field of complex industrial process modeling and fault diagnosis.
Background
Modern industrial production scale is becoming huge and process complexity is increasing day by day, and in order to ensure the smooth operation of production process, improve production efficiency and product quality, it becomes very important to monitor the production process.
Based on this background, multivariate statistical methods (MSPM) have been widely used in the field of process monitoring. The common multivariate statistical process monitoring methods include Principal Component Analysis (PCA), Partial Least Squares (PLS) and Independent Component Analysis (ICA). The PCA method is the most commonly used algorithm in the field of fault monitoring, can reduce the dimension of data, eliminate the correlation among variables, and monitor the process by establishing the statistics of principal component subspace and residual error subspace.
The traditional method for selecting the number of the principal elements comprises a cumulative variance contribution degree method (CPV), a reconstruction error variance method (VRE), a fault signal-to-noise ratio method (SNR) and the like. The CPV method mainly retains the principal component with the largest variance variation to a preset ratio (generally 85%), so that most information of the data can be considered to be contained in the principal component model, but there is no objective method for determining the ratio value in the method. The VRE method is to select pivot elements based on reconstruction errors, and when the reconstruction errors are minimum, the optimal number of the pivot elements is selected. The two methods are based on the ordering of the pivot elements based on the maximization of the variance, and the optimal pivot element number is selected from the aspect of dimension reduction, and the fault signal-to-noise ratio method considers the influence of the fault on the pivot element number, and selects the optimal pivot element number when the fault signal-to-noise ratio is the maximum.
In the field of fault monitoring, the selection of the number of the principal elements has very important influence on the monitoring performance. The Wanhaiqing and the like comprehensively consider the detection requirements of the number of the principal elements on different faults, and propose to adopt the optimal critical fault amplitude to determine the number of the principal elements and improve the monitoring capability on the faults; masayuki and the like determine the relation between the fault signal-to-noise ratio and the number of the pivot elements by utilizing prior fault information, and select the number of the pivot elements when the fault of the sensor has the maximum sensitivity; on the basis of a fault signal-to-noise ratio, Xuan and the like provide a concept of minimum detectable fault amplitude (MDFM), the MDFM is utilized to define a performance index of a detectable fault coverage range, and the number of the principal elements when the detectable fault coverage rate is maximum is the optimal number of the principal elements; prieto et al utilize Discriminant Analysis (DA) to select principal elements, improving the separability from class to the maximum extent, so as to achieve a better fault diagnosis effect. The method selects the optimal number of the pivot elements by defining a certain index and establishing the relationship between the fault monitoring effect and the number of the pivot elements so as to improve the fault monitoring precision. However, the principal element with a large variance change does not necessarily contain more fault information, and if the principal element with a large variance is simply placed in a subspace for monitoring, important fault information may be lost, so that there is a certain limitation in optimizing the number of the principal elements. The major elements which are more relevant to the fault are extracted and monitored through the Relief algorithm, so that the condition that the traditional PCA algorithm is in the major state is avoidedBlindness and subjectivity in meta selection; zhao et al propose a fault-related principal component analysis algorithm (FPCA), further divide principal component subspace and residual subspace into four subspaces, namely a fault-related subspace and a fault-unrelated subspace, for monitoring by using fault data information; Info-PCA method proposed by Grumtao et al, by constructing the accumulated T2And the change rate of the statistics is used for measuring the enrichment degree of the process information on each principal element, and the principal element direction with larger change rate is considered to be more important in fault detection, so that fault-related principal elements are extracted. Jiang et al propose a fault monitoring algorithm for fault-Sensitive Principal Components (SPCA) by monitoring T in real time2And selecting a plurality of pivot elements with the maximum change rate at the current moment for monitoring the change rate of the statistic. Wherein, the first two methods need the support of the fault data set when modeling, and the second two methods need to observe T2The rate of change of the statistics in each direction is used to pick the pivot. Therefore, in the PCA fault monitoring model, the selection of the pivot is very important.
Disclosure of Invention
Aiming at the problem of how to select pivot elements in the traditional PCA fault monitoring algorithm, a multi-block PCA fault monitoring method based on fault sensitive pivot element selection is provided.
A fault sensitive principal component selection-based multi-block PCA fault monitoring method comprises the following steps:
step 1: acquiring an original normal working condition data set, and carrying out standardization processing on the original normal working condition data set to obtain a new data set;
step 2: carrying out PCA decomposition on the data set X;
and step 3: defining the fault sensitivity coefficient of the ith pivot element to the jth variable, and dividing each sort result into a sub-block;
and 4, step 4: determining a threshold value epsilon for the sensitivity coefficient epsilonlimIn the jth sub-block, epsilon is selectedij＞εlimFront k of (2)jIndividual principal element carries out monitoring, kjI.e. the number of the selected principal elements in the jth sub-block, to obtain the T of the sub-block2Statistical quantity control limit
Figure BDA0001859916620000021
And 5: for a sample x to be monitored newly acquired from a sensortestSequentially calculating T in jth sub-block2Statistics;
step 6: after the monitoring results of all the subblocks are obtained, based on a Bayesian inference method, the monitoring results of all the subblocks are fused into a BIC statistic to obtain a final monitoring result, and when the BIC statistic exceeds a control limit, the monitoring sample is considered to have a fault.
Aiming at the problem of how to select the principal elements in the traditional PCA fault monitoring algorithm, the invention defines a fault sensitivity coefficient epsilon as a new principal element sorting criterion, and divides each sorting result into a sub-block based on m-dimension variables having m sorting results for the principal elements. Selecting the fault sensitivity coefficient epsilon larger than the threshold epsilon in each sub-blocklimThe principal element of (1) carries out fault monitoring and calculates T of each sub-block2Statistics are obtained. And fusing the monitoring results of the sub-blocks by using a Bayesian inference method to obtain a final BIC monitoring result.
According to the method, a fault sensitivity coefficient epsilon is defined as a new principal element sorting criterion, principal elements are sorted according to the sensitivity degree of faults occurring on each variable from large to small, and the positions of the faults occurring are unknown, so that m-dimensional variables have m sorting results. Calculating corresponding T by establishing sub-block PCA model for each sort result2And (5) statistics, establishing m sub-models together, and obtaining sub-block monitoring results. And then fusing the monitoring results of the sub-blocks based on a Bayesian method to obtain BIC statistics and making a final decision.
According to the method, on one hand, principal elements can be extracted without the aid of a fault data set, and on the other hand, computing resources consumed by real-time modeling are avoided.
Drawings
FIG. 1 is a diagram illustrating the effect of fault distribution on statistical monitoring.
Fig. 2 is a flow chart of a fault monitoring method of the present invention.
Fig. 3 is a schematic diagram showing comparison of monitoring results of a plurality of PCA (mbpsca) methods for numerical simulation fault 1 respectively adopting a PCA method and fault-sensitive principal component selection.
Fig. 4 is a schematic diagram showing comparison of monitoring results of a plurality of PCA (mbpsca) methods for numerical simulation fault 2 respectively adopting a PCA method and fault-sensitive principal component selection.
Fig. 5 is a monitoring scatter diagram of the numerical simulation fault 1 in each principal component direction.
Fig. 6 is a monitoring scatter diagram of the numerical simulation fault 2 in each principal component direction.
Figure 7 is a comparative schematic of the results of monitoring TE process fault 10.
FIG. 8 is a graphical comparison of the results of monitoring TE process fault 16.
Detailed Description
The invention will be described in further detail below with reference to fig. 2:
take a common chemical process-TE process and a numerical example as an example. Two faults set in the numerical example and 21 faults of the TE process are monitored. The TE process is a simulation system proposed by the Tenessee Eastman chemical company based on a certain actual chemical production process, and in the research in the field of process system engineering, the TE process is a common standard problem (Benchmark recipe) that better simulates many typical characteristics of an actual complex industrial process system, and thus is widely applied to the research of control, optimization, process monitoring and fault diagnosis as a simulation example. The TE process consists mainly of five main units, a reactor, a condenser, a compressor, a separator and a stripper. The process contains 22 process measurement variables, 19 constituent measurement variables and 12 manipulated variables. 22 process measurement variables and 11 manipulated variables outside the stirring speed were selected for modeling and monitoring. The TE process comprises 21 faults in total, 960 samples under normal working conditions are collected as a training data set, 960 samples under various fault working conditions are used as a fault test set, and faults are detected from a 161 st sample point
Step 1: obtaining an original normal condition data set X0∈Rn×mAnd carrying out standardization processing on the data to obtain a data set X epsilon Rn ×m. Wherein n represents the number of samples and m represents the number of variables. Standardization processing methodComprises the following steps:
Figure BDA0001859916620000041
wherein x0bRepresenting a data set X0The b-th sample of (1), xbRepresents the normalized b-th sample, mean (X)0) Representation matrix X0Mean vector of (1), std (X)0) Representation matrix X0The standard deviation vector of (2).
Step 2: the data set X is subjected to PCA decomposition.
Figure BDA0001859916620000051
T＝XP (3)
X＝TPT+E (4)
Where V is the covariance matrix
Figure BDA0001859916620000052
XTAnd X is used as an eigenvector matrix obtained by eigenvalue decomposition, and Λ is a diagonal matrix, wherein the elements on the diagonal are eigenvalues arranged from large to small. T is belonged to Rn×kRepresents a scoring matrix, P ∈ Rm×kRepresents the load matrix, which consists of the first k elements of matrix V, where k represents the number of principal elements selected in the PCA. E then represents the information in the residual space.
And step 3: defining the fault sensitivity coefficient epsilon of ith pivot element to jth variableijIs composed of
Figure BDA0001859916620000053
Wherein p isijIs an element of the ith row and jth column in the matrix V, λiIs the eigenvalue corresponding to the ith eigenvector. ε calculated by the equation (4)ijAnd according to epsilonijIs obtained by sorting the load vectors according to the size of the load vector
Figure BDA0001859916620000054
And m sequencing results are provided, wherein the superscript j represents the jth result and represents the size of the fault sensitivity coefficient of each pivot element on the jth dimension variable. Each sort result is divided into sub-blocks.
And 4, step 4: determining a threshold value epsilon for the sensitivity coefficient epsilonlimIn the jth sub-block, epsilon is selectedij＞εlimFront k of (2)jIndividual principal element carries out monitoring, kjI.e. the number of the selected pivot element in the jth sub-block, the T of the sub-block2Statistical quantity control limit
Figure BDA0001859916620000055
Is composed of
Figure BDA0001859916620000056
Where n represents the number of samples in the data set X,
Figure BDA0001859916620000057
denotes that F is distributed in a degree of freedom of kjAnd n-kjNext, the confidence is α, and usually α is 0.99.
And 5: for newly acquired sample x to be monitoredtestSequentially calculating T in jth sub-block2Statistics
Figure BDA0001859916620000058
Figure BDA0001859916620000059
Wherein
Figure BDA00018599166200000510
As a load vector
Figure BDA00018599166200000511
The corresponding characteristic value. If it is
Figure BDA00018599166200000512
Then it is considered asA failure occurred in j subblocks.
Step 6: after the monitoring results of all the subblocks are obtained, based on a Bayesian inference method, the monitoring results of all the subblocks are fused into BIC statistics to obtain a final monitoring result. The Bayesian inference method comprises the following steps:
in Bayesian inference, sample x is testedtestIn the jth sub-block T2The fault condition probability of a statistic can be expressed as:
Figure BDA0001859916620000061
Figure BDA0001859916620000062
wherein xtest,jRepresenting test samples in the jth sub-block, likelihood functions
Figure BDA0001859916620000063
And
Figure BDA0001859916620000064
the definition is as follows:
Figure BDA0001859916620000065
where "N" and "F" represent normal and fault conditions respectively,
Figure BDA0001859916620000066
is the prior probability of a normal sample, with a confidence of β
Figure BDA0001859916620000067
Is 1 to β;
Figure BDA0001859916620000068
is T of new sample in jth sub-block2Statistics;
Figure BDA0001859916620000069
is T of jth sub-block2A statistical quantity control limit. The final fused BIC statistic may be calculated by equation (10).
Figure BDA00018599166200000610
The BIC statistic control limit is 1- β when the BIC statistic exceeds the control limit, the monitoring sample is considered to be faulty.
FIG. 1 shows the effect of fault distribution on statistical monitoring, and the importance of selecting the number of principal elements can be seen.
Fig. 3 is a schematic diagram showing comparison of monitoring results of the numerical simulation fault 1 by using the PCA method and the MBSPCA method, respectively. The monitoring performance of the MBSPCA method in the fault 1 is far better than that of the PCA method
Fig. 4 is a schematic diagram showing comparison of monitoring results of the numerical simulation fault 2 by using the PCA method and the MBSPCA method, respectively. The monitoring performance of the MBSPCA method in the fault 2 is far better than that of the PCA method
Fig. 5 is a monitoring scatter diagram of the numerical simulation fault 1 in each principal component direction. It can be seen that only the faulty sample and the normal sample in the direction of the fifth principal element can be well distinguished.
Fig. 6 is a monitoring scatter diagram of the numerical simulation fault 2 in each principal component direction. It can be seen that the fault monitoring effect in the direction of the fourth principal element is the best
FIGS. 7 and 8 are schematic diagrams comparing the results of monitoring TE process faults 10 and 16, wherein the solid red line in the sub-graphs a and b is the control limit of the fault, the values are 1- β, and the curve is for each sample
Figure BDA0001859916620000071
Statistics, from T of each sub-block2The statistics are obtained by fusing the following equations (11). It can be seen that the monitoring effect of MBSPCA is far better than that of PCA. Sub-graph c is the monitoring effect when different numbers of principal elements are selected in the traditional PCA method, and can be seen as T2Several principal elements with larger contribution in the statistic are all located at the later position, and the principal elements are arranged in the traditional principal element selection methodIn addition, the monitoring effect of the fault is affected. And sub-graph d is T of each principal element in fault-related variable sub-block monitoring in the MBSPCA method2The statistic contribution degree shows that T of the principal element in the sub-block can be found2The contribution degrees are arranged from large to small, and the selected pivot elements can be ensured to contain sufficient fault information by selecting a plurality of pivot elements for monitoring, so that the fault monitoring is facilitated.

Claims (7)

1. A fault sensitive principal component selection-based multi-block PCA fault monitoring method is applied to a chemical process-TE process and used for monitoring faults in the TE process, and is characterized by comprising the following steps:
step 1: acquiring an original normal working condition data set, and carrying out standardization processing on the original normal working condition data set to obtain a new data set;
step 2: carrying out PCA decomposition on the data set X;
and step 3: defining the fault sensitivity coefficient of the ith pivot element to the jth variable, and dividing each sort result into a sub-block;
and 4, step 4: determining a threshold value epsilon for the sensitivity coefficient epsilonlimIn the jth sub-block, epsilon is selectedij＞εlimFront k of (2)jIndividual principal element carries out monitoring, kjI.e. the number of the selected principal elements in the jth sub-block, to obtain the T of the sub-block2Statistical quantity control limit
Figure FDA0002380423270000011
And 5: for a sample x to be monitored acquired by a sensortestSequentially calculating T in jth sub-block2Statistics;
step 6: after the monitoring results of all the subblocks are obtained, based on a Bayesian inference method, the monitoring results of all the subblocks are fused into a BIC statistic to obtain a final monitoring result, and when the BIC statistic exceeds a control limit, the monitoring sample is considered to have a fault.
2. The method for fault-sensitive principal component selection-based multiple PCA fault monitoring as claimed in claim 1, wherein the step 1 is:
obtaining an original normal condition data set X0∈Rn×mAnd carrying out standardization processing on the data to obtain a data set X epsilon Rn×mWherein n represents the number of samples, and m represents the number of variables; the standardization processing method comprises the following steps:
Figure FDA0002380423270000012
wherein x0bRepresenting a data set X0The b-th sample of (1), xbRepresents the normalized b-th sample, mean (X)0) Representation matrix X0Mean vector of (1), std (X)0) Representation matrix X0The standard deviation vector of (2).
3. The method for fault-sensitive principal component selection-based multiple PCA fault monitoring as claimed in claim 1, wherein the step 2 is:
PCA decomposition of dataset X:
Figure FDA0002380423270000021
T＝XP (3)
X＝TPT+E (4)
where V is the covariance matrix
Figure FDA0002380423270000022
The characteristic vector matrix obtained by characteristic value decomposition is carried out, Λ is a diagonal matrix, the elements on the diagonal are characteristic values arranged from large to small, and T belongs to Rn×kRepresents a scoring matrix, P ∈ Rm×kAnd representing a load matrix which is composed of the first k components of the matrix V, wherein k represents the number of the principal elements selected in the PCA, and E represents the information in the residual error space.
4. The fault-sensitive principal component selection-based multi-block PCA fault monitor of claim 1The measuring method is characterized in that the step 3 is as follows: defining the fault sensitivity coefficient epsilon of ith pivot element to jth variableijIs composed of
Figure FDA0002380423270000023
Wherein p isijIs an element of the ith row and jth column in the matrix V, λiIs the characteristic value corresponding to the ith characteristic vector, and the fault sensitivity coefficient epsilon is calculated by using the formula (4)ijAnd according to the fault susceptibility coefficient epsilonijIs obtained by sorting the load vectors according to the size of the load vector
Figure FDA0002380423270000024
The total m sequencing results are obtained, wherein the superscript j represents the jth result and represents the size of the fault sensitivity coefficient of each pivot element on the jth dimension variable; each sort result is divided into sub-blocks.
5. The method for fault-sensitive principal component selection-based multiple PCA fault monitoring as claimed in claim 1, wherein the step 4 is:
determining a threshold value epsilon for the sensitivity coefficient epsilonlimIn the jth sub-block, epsilon is selectedij＞εlimFront k of (2)jIndividual principal element carries out monitoring, kjI.e. the number of the selected pivot element in the jth sub-block, the T of the sub-block2Statistical quantity control limit
Figure FDA0002380423270000025
Comprises the following steps:
Figure FDA0002380423270000026
where n represents the number of samples in the data set X,
Figure FDA0002380423270000027
denotes that F is distributed in a degree of freedom of kjAnd n-kjIn the following, the first and second parts of the material,the confidence level is α, typically α is 0.99.
6. The method for fault-sensitive principal component selection-based multiple PCA fault monitoring as claimed in claim 1, wherein the step 5 is:
for newly acquired sample x to be monitoredtestSequentially calculating T in jth sub-block2Statistics
Figure FDA0002380423270000031
Figure FDA0002380423270000032
Wherein
Figure FDA0002380423270000033
As a load vector
Figure FDA0002380423270000034
The corresponding characteristic value; if it is
Figure FDA0002380423270000035
It is assumed that a failure has occurred in the jth sub-block.
7. The method for fault-sensitive principal component selection-based multiple PCA fault monitoring as claimed in claim 1, wherein the step 6 is:
after the monitoring results of all the subblocks are obtained, fusing the monitoring results of all the subblocks into BIC statistic on the basis of a Bayesian inference method to obtain a final monitoring result; the Bayesian inference method comprises the following steps:
in Bayesian inference, a newly acquired sample x to be monitoredtestIn the jth sub-block T2The fault condition probability of a statistic can be expressed as:
Figure FDA0002380423270000036
Figure FDA0002380423270000037
wherein xtest,jRepresenting test samples in the jth sub-block, likelihood functions
Figure FDA0002380423270000038
And
Figure FDA0002380423270000039
the definition is as follows:
Figure FDA00023804232700000310
where "N" and "F" represent normal and fault conditions respectively,
Figure FDA00023804232700000311
is the prior probability of a normal sample, with a confidence of β
Figure FDA00023804232700000312
Is 1 to β;
Figure FDA00023804232700000313
is T of new sample in jth sub-block2Statistics;
Figure FDA00023804232700000314
is T of jth sub-block2A statistical quantity control limit; the final fused BIC statistic may be calculated by equation (10):
Figure FDA00023804232700000315
the control limit of the BIC statistic is 1- β, and when the BIC statistic exceeds the control limit, the monitoring sample is considered to be in fault.
CN201811330777.9A 2018-11-09 2018-11-09 Multi-block PCA fault monitoring method based on fault sensitive principal component selection Active CN109240276B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811330777.9A CN109240276B (en) 2018-11-09 2018-11-09 Multi-block PCA fault monitoring method based on fault sensitive principal component selection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811330777.9A CN109240276B (en) 2018-11-09 2018-11-09 Multi-block PCA fault monitoring method based on fault sensitive principal component selection

Publications (2)

Publication Number Publication Date
CN109240276A CN109240276A (en) 2019-01-18
CN109240276B true CN109240276B (en) 2020-06-05

Family

ID=65077783

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811330777.9A Active CN109240276B (en) 2018-11-09 2018-11-09 Multi-block PCA fault monitoring method based on fault sensitive principal component selection

Country Status (1)

Country Link
CN (1) CN109240276B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111913462B (en) * 2019-09-07 2022-03-18 宁波大学 Chemical fault monitoring method based on generalized multiple independent element analysis model
CN112184034B (en) * 2020-09-30 2024-04-09 江南大学 Multi-block k-nearest neighbor fault monitoring method and system based on mutual information
CN112611970A (en) * 2020-11-25 2021-04-06 南京中车浦镇海泰制动设备有限公司 Electromechanical clamp fault detection method based on PCA statistics
CN115291582A (en) * 2022-02-22 2022-11-04 江南大学 Method, device and equipment for monitoring faults in iron-making process

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101403923A (en) * 2008-10-31 2009-04-08 浙江大学 Course monitoring method based on non-gauss component extraction and support vector description
CN101477372A (en) * 2009-01-07 2009-07-08 沈阳化工学院 Fault separation technique for chemical production abnormal sub-domain
US10372120B2 (en) * 2016-10-06 2019-08-06 General Electric Company Multi-layer anomaly detection framework
CN108664009B (en) * 2017-08-03 2021-05-25 湖州师范学院 Stage division and fault detection method based on correlation analysis
CN108664402B (en) * 2018-05-14 2020-05-26 北京航空航天大学 Fault prediction method based on software network feature learning
CN108758969B (en) * 2018-06-14 2020-09-15 河南科技大学 Fault detection method and system for water chilling unit

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Blind Source Separation based on Principal Component Analysis- Independent Component Analysis for Acoustic Signal during Laser Welding Process;Ao Sansan;《 2010 International Conference on Digital Manufacturing & Automation 》;20110124;全文 *
Induction machine faults diagnosis by statistical neural networks with selection variables based on Principal component analysis;Sameh Marmouch;《18th international conference on Sciences and Techniques of Automatic control &computer engineering》;20171223;全文 *
Principal Component Analysis Preprocessing with Bayesian Networks for Battery Capacity Estimation;Liessman E. Sturlaugson;《2013 IEEE International Instrumentation and Measurement Technology Conference》;20130715;全文 *
基于多块信息提取的PCA故障诊断方法;顾炳斌，熊伟丽;《化工学报》;20181204;全文 *

Also Published As

Publication number Publication date
CN109240276A (en) 2019-01-18

Similar Documents

Publication Publication Date Title
CN109240276B (en) Multi-block PCA fault monitoring method based on fault sensitive principal component selection
CN111222290B (en) Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment
CN106355030B (en) A kind of fault detection method based on analytic hierarchy process (AHP) and Nearest Neighbor with Weighted Voting Decision fusion
CN116757534B (en) Intelligent refrigerator reliability analysis method based on neural training network
CN106845526B (en) A kind of relevant parameter Fault Classification based on the analysis of big data Fusion of Clustering
CN109409425B (en) Fault type identification method based on neighbor component analysis
CN112284440B (en) Sensor data deviation self-adaptive correction method
CN111191726B (en) Fault classification method based on weak supervision learning multilayer perceptron
CN113642754A (en) Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network
CN112200104A (en) Chemical engineering fault diagnosis method based on novel Bayesian framework for enhanced principal component analysis
CN112414694A (en) Equipment multistage abnormal state identification method and device based on multivariate state estimation technology
CN112000081B (en) Fault monitoring method and system based on multi-block information extraction and Mahalanobis distance
US20050144537A1 (en) Method to use a receiver operator characteristics curve for model comparison in machine condition monitoring
CN110782546A (en) Resistivity virtual measurement method of semiconductor PVD (physical vapor deposition) process based on combined tree model
CN113420061A (en) Steady state working condition analysis method, optimization debugging method and system of oil refining and chemical production device
CN115373370A (en) Method and system for monitoring running state of programmable controller
CN112433928A (en) Fault prediction method, device, equipment and storage medium of storage equipment
CN114429238A (en) Wind turbine generator fault early warning method based on space-time feature extraction
Cudney et al. A comparison of the Mahalanobis-Taguchi system to a standard statistical method for defect detection
CN109389313B (en) Fault classification diagnosis method based on weighted neighbor decision
CN112380763A (en) System and method for analyzing reliability of in-pile component based on data mining
TWI639908B (en) Method for detecting and diagnosing an abnormal process
Buschmann et al. Data-driven decision support for process quality improvements
CN116204825A (en) Production line equipment fault detection method based on data driving
CN115718880A (en) Prediction method for degradation stage of complex equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant