CN114034332A - Fault monitoring method for weapon equipment system - Google Patents

Fault monitoring method for weapon equipment system Download PDF

Info

Publication number
CN114034332A
CN114034332A CN202111394862.3A CN202111394862A CN114034332A CN 114034332 A CN114034332 A CN 114034332A CN 202111394862 A CN202111394862 A CN 202111394862A CN 114034332 A CN114034332 A CN 114034332A
Authority
CN
China
Prior art keywords
matrix
input
output
data
sensing data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111394862.3A
Other languages
Chinese (zh)
Other versions
CN114034332B (en
Inventor
胡昌华
孔祥玉
罗家宇
李红增
陈雅琳
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.)
Rocket Force University of Engineering of PLA
Original Assignee
Rocket Force University of Engineering of PLA
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 Rocket Force University of Engineering of PLA filed Critical Rocket Force University of Engineering of PLA
Priority to CN202111394862.3A priority Critical patent/CN114034332B/en
Publication of CN114034332A publication Critical patent/CN114034332A/en
Application granted granted Critical
Publication of CN114034332B publication Critical patent/CN114034332B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

The invention discloses a fault monitoring method for a weapon equipment system, which comprises the following steps: s1, collecting sensing data of the weapon equipment system, and carrying out linear standardization processing on the sensing data to obtain an input standard sensing data set and an output standard sensing data set; s2, performing high-dimensional mapping and nonlinear processing on the input standard sensing data set and the output standard sensing data set to obtain a nonlinear input matrix, a nonlinear output matrix, an input parameter matrix and an output parameter matrix; s3, calculating an input score matrix, an output score matrix, an input load matrix and an output load matrix; s4, collecting sensing data of the weapon equipment system to be detected to obtain sensing data to be detected, and carrying out fault monitoring on the sensing data to be detected according to the input score matrix, the output score matrix, the input load matrix and the output load matrix; the invention solves the problem that the fault judgment is inaccurate in the existing fault monitoring method for the weapon equipment system.

Description

Fault monitoring method for weapon equipment system
Technical Field
The invention relates to the field of weapon equipment system quality monitoring, in particular to a fault monitoring method for a weapon equipment system.
Background
With the development of modern technology, the system structure of large complex equipment such as rockets and missiles is more complex and cannot be described by simple physical models and mechanism models, so that a series of fault diagnosis methods based on fault trees and physical models are difficult to apply. With the development of integrated circuits, a large number of sensors are placed in many critical locations for equipment testing, while obtaining a large amount of data. Therefore, the method based on data-driven modeling is gradually becoming a research hotspot and is greatly developed and applied. Data acquired in an industrial process usually has the characteristics of high coupling, non-gaussian, non-linear and the like, and how to put forward a process characteristic building model from the data is a difficult point of a data driving method. Meanwhile, when the performance of the equipment is degraded, parameters drift in the test process and even a certain link fails, abnormal changes occur in process data monitored by a corresponding sensor, and how to find out the cause of the failure from the changes is a key and difficult point of current research.
In a non-linear process, the measured values of the process variables typically vary non-linearly, and there is often a non-linear relationship between the process variables. For the process, the traditional multivariate statistical analysis method applied to the stable process is difficult to effectively monitor the process, and larger false alarm and alarm missing conditions exist. Meanwhile, in the nonlinear process monitoring considering the key performance indexes, the key index variables also change in a nonlinear manner, and the traditional nonlinear partial least square method only considers that the process variables change in a nonlinear manner, so that the traditional nonlinear partial least square method is difficult to be well applied to the input and output nonlinear process. In order to study the relation between nonlinear input and nonlinear output and extract nonlinear characteristics, an input-output nonlinear model is provided. Based on the constructed model, the non-linear quality-dependent and quality-independent subspaces are partitioned and statistics are established in the respective subspaces. An online nonlinear process monitoring strategy is also designed, and comprises the steps of constructing a control limit offline, mapping projection and constructing statistic in an online process, and realizing real-time monitoring of online samples.
Disclosure of Invention
Aiming at the defects in the prior art, the fault monitoring method for the weapon equipment system provided by the invention solves the problem that the fault judgment of the existing fault monitoring method for the weapon equipment system is inaccurate.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a fault monitoring method for a weaponry system, comprising the steps of:
s1, collecting sensing data of the weapon equipment system, and carrying out linear standardization processing on the sensing data to obtain an input standard sensing data set and an output standard sensing data set;
s2, performing high-dimensional mapping and nonlinear processing on the input standard sensing data set and the output standard sensing data set to obtain a nonlinear input matrix, a nonlinear output matrix, an input parameter matrix and an output parameter matrix;
s3, calculating an input score matrix, an output score matrix, an input load matrix and an output load matrix according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix;
s4, collecting sensing data of the weapon equipment system to be detected to obtain the sensing data to be detected, and carrying out fault monitoring on the sensing data to be detected according to the input score matrix, the output score matrix, the input load matrix and the output load matrix.
Further, the types of the sensing data in step S1 include: voltage current data, pneumatic hydraulic data, pressure data, vibration data, temperature data, servo speed data, servo rotational speed data, and servo feedback voltage data, wherein the voltage current data, the pneumatic hydraulic data, the pressure data, the vibration data, and the temperature data are used as an input data set X, and the servo speed data, the servo rotational speed data, and the servo feedback voltage data are used as an output data set Y.
Further, the formula of the linear normalization process performed on the sensed data in step S1 is:
Figure BDA0003369670880000031
Figure BDA0003369670880000032
wherein ,
Figure BDA0003369670880000033
standard sensing data of j-th sampling sensing data for corresponding ith variable in input data set Xi，jSampling sensed data for the ith variable in the input data set X for the jth timei，jThe sensed data is sampled for the jth variable in the output data set Y,
Figure BDA0003369670880000034
the standard sensing data of j-th sampling sensing data of the corresponding ith variable in the output data set Y, n is the sampling times of each variable,
Figure BDA0003369670880000035
the constructed set is an input standard sensing dataset,
Figure BDA0003369670880000036
the constructed set is an output standard sensing dataset.
The beneficial effects of the above further scheme are: the data of different scales measured by different sensors are unified in dimension and scale, information loss in the data dimension reduction process is avoided, and abnormal representation of fault data statistics in the fault detection of the complex equipment system is further ensured (the abnormal representation is abnormal change of the fault data dimension reduction structure statistics compared with normal data statistics).
Further, step S2 includes the following substeps:
s21, mapping the input standard sensing data set and the output standard sensing data set to a high-dimensional space respectively to obtain a high-dimensional input matrix phixAnd high dimensional output matrix phiy；
S22, inputting the matrix phi to the high dimensionxAnd high dimensional outputMatrix phiyPerforming nonlinear processing to obtain nonlinear input matrix
Figure BDA0003369670880000037
And a non-linear output matrix
Figure BDA0003369670880000038
S23, according to the nonlinear input matrix
Figure BDA0003369670880000039
And a non-linear output matrix
Figure BDA00033696708800000310
Calculating an input parameter matrix
Figure BDA00033696708800000311
And output parameter matrix
Figure BDA00033696708800000312
Further, the nonlinear input matrix in step S22
Figure BDA00033696708800000313
And a non-linear output matrix
Figure BDA00033696708800000314
The calculation formula of (2) is as follows:
Figure BDA00033696708800000315
Figure BDA0003369670880000041
calculating an input parameter matrix in the step S23
Figure BDA0003369670880000042
And output parameter matrix
Figure BDA0003369670880000043
The formula of (1) is:
Figure BDA0003369670880000044
Figure BDA0003369670880000045
wherein ,1nIs composed of
Figure BDA0003369670880000046
The vector of all 1 columns of the image,
Figure BDA0003369670880000047
is a real space.
The beneficial effects of the above further scheme are: the original data monitored in the nonlinear process of the complex equipment is mapped to a high dimension, so that the nonlinear data can construct a linear relation in a high dimension space, and meanwhile, the kernel function is constructed to avoid the specific construction of the nonlinear function, thereby reducing the computational complexity of the nonlinear process and improving the modeling efficiency of fault detection.
Further, step S3 includes the following substeps:
s31, constructing a nonlinear feature extraction target function according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix, and solving the nonlinear feature extraction target function by adopting Lagrange' S theorem to obtain an input-output parameter relation equation;
s32, calculating initial output latent variable component information according to the input and output parameter relation equation;
and S33, extracting an input score matrix, an output score matrix, an input load matrix and an output load matrix by a nonlinear regression iteration method according to the initial output latent variable component information.
Further, the nonlinear feature extraction objective function in step S31 is:
Figure BDA0003369670880000048
the input and output parameter relation equation is as follows:
Figure BDA0003369670880000049
wherein w is a high-dimensional input matrix phixC is a high-dimensional output matrix phiyThe vector of the projection of (a) is,
Figure BDA00033696708800000410
in the case of a non-linear input matrix,
Figure BDA00033696708800000411
in the form of a non-linear output matrix,
Figure BDA00033696708800000412
in order to input the parameter matrix, the parameter matrix is,
Figure BDA00033696708800000413
and u is the initial output latent variable component information, and theta is the characteristic value corresponding to u after characteristic decomposition.
The beneficial effects of the above further scheme are: the modeling process of the equipment testing system with input and output in nonlinear change is given out overall, the objective function of input and output correlation extraction is established, and the dimension reduction of the equipment testing data is realized by extracting the projection direction.
Further, step S33 includes the following substeps:
s331, initializing the number of iterations i to 1, and outputting latent variable component information uiAssigning an initial value u;
s332, calculating input latent variable component information t of the ith iterationiAnd updating output latent variable component information ui；
Figure BDA0003369670880000051
Figure BDA0003369670880000052
wherein ,u′iFor updated output latent variable component signali；
S333, from output latent variable component information u'iAnd input latent variable component information tiCalculating input load vector and output load vector, and applying to nonlinear input matrix
Figure BDA0003369670880000053
Non-linear output matrix
Figure BDA0003369670880000054
Input parameter matrix
Figure BDA0003369670880000055
And output parameter matrix
Figure BDA0003369670880000056
Updating is carried out;
Figure BDA0003369670880000057
Figure BDA0003369670880000058
Figure BDA0003369670880000059
Figure BDA00033696708800000510
Figure BDA00033696708800000511
Figure BDA00033696708800000512
wherein ,py，iTo output the load vector, px，iFor the input load vector, E is the identity matrix,
Figure BDA00033696708800000513
for updated non-linear input matrix
Figure BDA00033696708800000514
For updated non-linear output matrix
Figure BDA00033696708800000515
For updated input parameter matrix
Figure BDA00033696708800000516
For updated output parameter matrix
Figure BDA00033696708800000517
S334, pair
Figure BDA0003369670880000061
In
Figure BDA0003369670880000062
Performing characteristic decomposition to obtain output latent variable component information ui+1；
S335, judging and outputting latent variable component information ui+1If yes, jumping to step S336, and if no, adding 1 to the iteration number i, and jumping to step S332;
s336, extracting A pieces of output latent variable component information uiA pieces of input latent variable component information tiA number of input load vectors px，iAnd A output loadsVector py，iObtaining the input score matrix M ═ t (t)1，...，tA) And outputting a score matrix U ═ U1，...，uA) Input load matrix Px＝(px，1，...，px，A) And an output load matrix Py＝(py，1，...，py，A)。
The beneficial effects of the above further scheme are: and a nonlinear projection direction is specifically constructed by nonlinear iteration, and dimension reduction is realized by projecting along the direction after nonlinear mapping of test data in online monitoring.
Further, step S4 includes the following substeps:
s41, calculating latent variables of single historical sensing data after dimension reduction projection:
Figure BDA0003369670880000063
wherein t is latent variable after dimension reduction projection of single historical sensing data, M is an input score matrix,
Figure BDA0003369670880000064
the method comprises the steps of performing Gaussian kernel functions on single historical sensing data and all historical sensing data, and performing standardization;
s42, according to the latent variable after dimension reduction projection is carried out on the single historical sensing data, establishing the quality-independent spatial statistic of the historical sensing data:
Q＝||φ(xi)-Pxt||2
wherein Q is the quality independent spatial statistic of the historical sensing data, phi (x)i) To be xiFunction for high dimensional mapping, xi＝{xi1，...，xij，...，xim}，||||2Performing two-norm operation;
s43, calculating a quality-related spatial control limit of the historical sensing data:
Figure BDA0003369670880000065
wherein ,
Figure BDA0003369670880000066
quality-dependent spatial control limits for historical sensory data, FA，n-AIs an F distribution;
s44, calculating the quality-independent space control limit of the historical sensing data according to the quality-independent space statistic of the historical sensing data:
Figure BDA0003369670880000071
wherein g is xi/2 mu, h is 2 mu2/ξ，Jth，SPEIs the quality-independent space control limit of the historical sensing data, g is a control limit coefficient, xi is the variance after all the historical sensing data construct a statistic Q, mu is the mean after all the historical sensing data construct the statistic Q,
Figure BDA0003369670880000072
is chi-square distribution, and h is the degree of freedom of chi-square distribution;
s45, collecting sensing data of the weapon equipment system to be tested to obtain the sensing data to be tested;
s46, mapping the to-be-detected sensing data after standardized processing to a high-dimensional space to obtain nonlinear to-be-detected sensing data;
s47, carrying out standardization processing on the Gaussian kernel function of the nonlinear sensing data to be measured to obtain the normalized Gaussian kernel function:
Figure BDA0003369670880000073
wherein ,
Figure BDA0003369670880000074
to normalize the processed Gaussian kernel function, knewIs a Gaussian kernel function of the nonlinear sensing data to be measured, and K isAn input data set X and a gaussian kernel function of the input data set X;
s48, projecting the nonlinear sensing data to be measured along the quality-related space according to the normalized Gaussian kernel function to obtain a latent variable tnew：
Figure BDA0003369670880000075
S49, calculating the quality-related spatial statistic and the quality-unrelated spatial statistic of the to-be-measured sensing data:
Figure BDA0003369670880000076
Figure BDA0003369670880000077
wherein ,T′2Is the quality-related spatial statistic of the sensed data to be measured, Q' is the quality-independent spatial statistic of the sensed data to be measured,
Figure BDA0003369670880000078
for the non-linear sensed data to be measured,
Figure BDA0003369670880000079
s50, judging quality related space statistic T 'of to-be-detected sensing data'2Whether or not to be greater than or equal to a quality-related spatial control limit of the historical sensory data
Figure BDA0003369670880000081
If yes, the existence of the weapon equipment system to be tested causes the fault of system operation error, namely, the existence of quality-related fault, if not, the step S51 is skipped to;
s51, judging whether the quality-independent space statistic Q' of the sensing data to be detected is more than or equal to the quality-independent space control limit J of the historical sensing datath，SPEIf, ifIf the measured data of the weapon equipment system to be measured is normal data, the weapon equipment system to be measured operates normally.
The beneficial effects of the above further scheme are: the control limit of fault detection in the test process of the equipment system is determined by historical sensing data, and the real-time monitoring of the test process is realized through the statistic construction of the sensing data to be detected.
In conclusion, the beneficial effects of the invention are as follows:
the existing method only considers the process that the input changes in a nonlinear way, so that the complex nonlinear process aimed by the method cannot be accurately described in the modeling and feature extraction stages, and the complex nonlinear process is finally reflected in higher fault detection rate and lower false alarm rate. The invention constructs the objective function of the correlation extraction of nonlinear input and output, constructs the projection direction of the quality correlation through nonlinear iteration, effectively extracts the nonlinear characteristics of the input and output data, has good process monitoring performance, and has more excellent quality correlation fault detection rate and lower false alarm rate. And the type and severity of the fault of the weapon equipment system can be judged according to the sensing data in the operation process of the weapon equipment system.
Drawings
FIG. 1 is a flow chart of a fault monitoring method for a weaponry system;
fig. 2 shows the fault detection result of the quality-related fault IDV (14);
fig. 3 shows the fault detection result of the quality-independent fault IDV (04).
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
As shown in fig. 1, a fault monitoring method for a weaponry system includes the steps of:
s1, collecting sensing data of the weapon equipment system, and carrying out linear standardization processing on the sensing data to obtain an input standard sensing data set and an output standard sensing data set;
the types of the sensed data in step S1 include: voltage current data, pneumatic hydraulic data, pressure data, vibration data, temperature data, servo speed data, servo rotational speed data, and servo feedback voltage data, wherein the voltage current data, the pneumatic hydraulic data, the pressure data, the vibration data, and the temperature data are used as an input data set X, and the servo speed data, the servo rotational speed data, and the servo feedback voltage data are used as an output data set Y.
Monitoring the corresponding type fault of the weaponry system, namely acquiring process data of the corresponding type fault to obtain an input data set X, taking output data which is output by the weaponry system and can reflect the corresponding fault as an output data set Y, for example, when the voltage and the current are increased, the rotating speed of a servo mechanism is increased, then the voltage and the current data are process data, namely input data, the rotating speed of the servo mechanism is taken as index data, namely output data, and the corresponding relation between the input data and the output data under normal conditions is established and can be used for monitoring the operating condition of the system.
The formula for performing linear normalization processing on the sensed data in step S1 is:
Figure BDA0003369670880000091
Figure BDA0003369670880000092
wherein ,
Figure BDA0003369670880000093
for the corresponding ith in the input data set XVariable j-th time standard sensing data x of sampled sensing datai，jSampling sensed data for the ith variable in the input data set X for the jth timei，jThe sensed data is sampled for the jth variable in the output data set Y,
Figure BDA0003369670880000101
the standard sensing data of j-th sampling sensing data of the corresponding ith variable in the output data set Y, n is the sampling times of each variable,
Figure BDA0003369670880000102
the constructed set is an input standard sensing dataset,
Figure BDA0003369670880000103
the constructed set is an output standard sensing dataset.
The above variables are understood to be a certain type of sensory data, e.g. xi，jWhich can be interpreted as the j-th sampled sensing data of the voltage data i in the input data set X.
S2, performing high-dimensional mapping and nonlinear processing on the input standard sensing data set and the output standard sensing data set to obtain a nonlinear input matrix, a nonlinear output matrix, an input parameter matrix and an output parameter matrix;
step S2 includes the following substeps:
s21, mapping the input standard sensing data set and the output standard sensing data set to a high-dimensional space respectively to obtain a high-dimensional input matrix phixAnd high dimensional output matrix phiy；
wherein ,Φx＝[φ(x1)，φ(x2)，...，φ(xn)]T，Φy＝[φ(y1)，φ(y2)，...，φ(yn)]T，，φ(xi) To be xiFunction for high dimensional mapping, xi＝{xi1，...，xij，...，xim}，yi＝{yi1，...，yij，...，yimAnd m is the number of variables.
S22, inputting the matrix phi to the high dimensionxAnd high dimensional output matrix phiyPerforming nonlinear processing to obtain nonlinear input matrix
Figure BDA0003369670880000104
And a non-linear output matrix
Figure BDA0003369670880000105
Nonlinear input matrix in step S22
Figure BDA0003369670880000106
And a non-linear output matrix
Figure BDA0003369670880000107
The calculation formula of (2) is as follows:
Figure BDA0003369670880000108
Figure BDA0003369670880000109
s23, according to the nonlinear input matrix
Figure BDA00033696708800001010
And a non-linear output matrix
Figure BDA00033696708800001011
Calculating an input parameter matrix
Figure BDA00033696708800001012
And output parameter matrix
Figure BDA00033696708800001013
Calculating an input parameter matrix in the step S23
Figure BDA00033696708800001014
And output parameter matrix
Figure BDA00033696708800001015
The formula of (1) is:
Figure BDA0003369670880000111
Figure BDA0003369670880000112
wherein ,1nIs composed of
Figure BDA0003369670880000113
The vector of all 1 columns of the image,
Figure BDA0003369670880000114
is a real space.
S3, calculating an input score matrix, an output score matrix, an input load matrix and an output load matrix according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix;
step S3 includes the following substeps:
s31, constructing a nonlinear feature extraction target function according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix, and solving the nonlinear feature extraction target function by adopting Lagrange' S theorem to obtain an input-output parameter relation equation;
the nonlinear feature extraction objective function in step S31 is:
Figure BDA0003369670880000115
the input and output parameter relation equation is as follows:
Figure BDA0003369670880000116
wherein w is a high-dimensional input matrix phixC is a high-dimensional output matrix phiyThe vector of the projection of (a) is,
Figure BDA0003369670880000117
in the case of a non-linear input matrix,
Figure BDA00033696708800001111
in the form of a non-linear output matrix,
Figure BDA0003369670880000118
in order to input the parameter matrix, the parameter matrix is,
Figure BDA0003369670880000119
for the output parameter matrix, u is the initial output latent variable component information, θ2For the characteristic value of the input-output parameter relation equation, θ ═ 2 λ ═ 2 β.
The detailed process of solving the nonlinear feature extraction target function by adopting the Lagrange theorem is as follows:
a1, establishing a Lagrange theorem solving formula:
Figure BDA00033696708800001110
where λ and β are parameters (fixed expressions) of the lagrangian multiplier method.
A2, partial derivatives of w and c in A1 are calculated:
Figure BDA0003369670880000121
a3, according to constraints: if w | | | c | | | 1, the expression in a2 can be simplified as:
Figure BDA0003369670880000122
a4, substituting the formula in step A3 into step A2:
Figure BDA0003369670880000123
a5, formula in simplified step A4:
Figure BDA0003369670880000124
a6 multiplication of both sides of the formula in step A5
Figure BDA0003369670880000125
Figure BDA0003369670880000126
Figure BDA0003369670880000127
u＝Φyc is
Figure BDA0003369670880000128
Latent variables (dimensionality reduction data, comprising
Figure BDA0003369670880000129
Most information of) of (ii), formula (ii)
Figure BDA00033696708800001210
Can be rewritten as:
Figure BDA00033696708800001211
s32, calculating initial output latent variable component information according to the input and output parameter relation equation;
in the step S32, the relation between the input and output parameters is determined
Figure BDA00033696708800001212
And (4) carrying out characteristic decomposition to obtain u.
And S33, extracting an input score matrix, an output score matrix, an input load matrix and an output load matrix by a nonlinear regression iteration method according to the initial output latent variable component information.
Step S33 includes the following substeps:
s331, initializing the number of iterations i to 1, and outputting latent variable component information uiAssigning an initial value u;
s332, calculating input latent variable component information t of the ith iterationiAnd updating output latent variable component information ui；
Figure BDA00033696708800001213
Figure BDA0003369670880000131
wherein ,U′iFor updated output latent variable component information ui；
S333, from output latent variable component information u'iAnd input latent variable component information tiCalculating input load vector and output load vector, and applying to nonlinear input matrix
Figure BDA0003369670880000132
Non-linear output matrix
Figure BDA0003369670880000133
Input parameter matrix
Figure BDA0003369670880000134
And output parameter matrix
Figure BDA0003369670880000135
Updating is carried out;
Figure BDA0003369670880000136
Figure BDA0003369670880000137
Figure BDA0003369670880000138
Figure BDA00033696708800001318
Figure BDA0003369670880000139
Figure BDA00033696708800001310
wherein ,py，iTo output the load vector, px，iFor the input load vector, E is the identity matrix,
Figure BDA00033696708800001311
for updated non-linear input matrix
Figure BDA00033696708800001312
For updated non-linear output matrix
Figure BDA00033696708800001313
For updated input parameter matrix
Figure BDA00033696708800001314
For updated output parameter matrix
Figure BDA00033696708800001315
S334, pair
Figure BDA00033696708800001316
In
Figure BDA00033696708800001317
Performing characteristic decomposition to obtain output latent variable component information ui+1；
S335, judging and outputting latent variable component information ui+1If yes, jumping to step S336, and if no, adding 1 to the iteration number i, and jumping to step S332;
s336, extracting A pieces of output latent variable component information uiA pieces of input latent variable component information tiA number of input load vectors px，iAnd A output load vectors py，iObtaining the input score matrix M ═ t (t)1，...，tA) And outputting a score matrix U ═ U1，...，uA) Input load matrix Px＝(px，1，...，px，A) And an output load matrix Py＝(py，1，...，py，A)。
S4, collecting sensing data of the weapon equipment system to be detected to obtain the sensing data to be detected, and carrying out fault monitoring on the sensing data to be detected according to the input score matrix, the output score matrix, the input load matrix and the output load matrix.
Step S4 includes the following substeps:
s41, calculating latent variables of single historical sensing data after dimension reduction projection:
Figure BDA0003369670880000141
wherein t is latent variable after dimension reduction projection of single historical sensing data, M is an input score matrix,
Figure BDA0003369670880000146
the method comprises the steps of performing Gaussian kernel functions on single historical sensing data and all historical sensing data, and performing standardization;
s42, according to the latent variable after dimension reduction projection is carried out on the single historical sensing data, establishing the quality-independent spatial statistic of the historical sensing data:
Q＝||φ(xi)-Pxt||2
wherein Q is the quality independent spatial statistic of the historical sensing data, phi (x)i) To be xiFunction for high dimensional mapping, xi＝{xi1，...，xij，...，xim}，||||2Performing two-norm operation;
s43, calculating a quality-related spatial control limit of the historical sensing data:
Figure BDA0003369670880000142
wherein ,
Figure BDA0003369670880000143
quality-dependent spatial control limits for historical sensory data, FA，n-AIs an F distribution;
s44, calculating the quality-independent space control limit of the historical sensing data according to the quality-independent space statistic of the historical sensing data:
Figure BDA0003369670880000144
wherein g is xi/2 mu, h is 2 mu2/ξ，Jth，SPEIs the quality-independent space control limit of the historical sensing data, g is a control limit coefficient, xi is the variance after all the historical sensing data construct a statistic Q, mu is the mean after all the historical sensing data construct the statistic Q,
Figure BDA0003369670880000145
is chi-square distribution, and h is the degree of freedom of chi-square distribution;
s45, collecting sensing data of the weapon equipment system to be tested to obtain the sensing data to be tested;
s46, mapping the to-be-detected sensing data after standardized processing to a high-dimensional space to obtain nonlinear to-be-detected sensing data;
s47, carrying out standardization processing on the Gaussian kernel function of the nonlinear sensing data to be measured to obtain the normalized Gaussian kernel function:
Figure BDA0003369670880000151
wherein ,
Figure BDA0003369670880000152
to normalize the processed Gaussian kernel function, knewThe method comprises the following steps that a Gaussian kernel function of nonlinear sensing data to be detected is adopted, and K is an input data set X and the Gaussian kernel function of the input data set X;
s48, projecting the nonlinear sensing data to be measured along the quality-related space according to the normalized Gaussian kernel function to obtain a latent variable tnew：
Figure BDA0003369670880000153
S49, calculating the quality-related spatial statistic and the quality-unrelated spatial statistic of the to-be-measured sensing data:
Figure BDA0003369670880000154
Figure BDA0003369670880000155
wherein ,T′2Is the quality-related spatial statistic of the sensed data to be measured, Q' is the quality-independent spatial statistic of the sensed data to be measured,
Figure BDA0003369670880000156
for the non-linear sensed data to be measured,
Figure BDA0003369670880000157
s50, judging quality related space statistic T 'of to-be-detected sensing data'2Whether or not to be greater than or equal to a quality-related spatial control limit of the historical sensory data
Figure BDA0003369670880000158
If yes, the existence of the weapon equipment system to be tested causes the fault of system operation error, namely, the existence of quality-related fault, if not, the step S51 is skipped to;
s51, judging whether the quality-independent space statistic Q' of the sensing data to be detected is more than or equal to the quality-independent space control limit J of the historical sensing datath，SPEIf the measured data of the weapon equipment system to be measured is normal data, the weapon equipment system to be measured operates normally.
Faults that cause the system to operate incorrectly: a large number of sensors for monitoring process variables are distributed in the equipment testing process, the sensing data of the sensors for monitoring the process variables are collected to obtain an input data set X, and if the process variables are abnormal due to a fault in the process, the key performance indexes are abnormal, namely the output data set Y is abnormal, the fault causes the operation error of the system;
disturbance which does not cause system operation error: similarly, if the process is failed and the process variable is abnormal, but the key performance index is not abnormal, the system operation is not disturbed.
Experiment:
considering that the characteristics of a large complex equipment system such as multiple indexes, high dimensionality, large samples, existence of process variables, key performance indexes and the like are similar to Tennessee-Islaman (TEP), the TEP is adopted to verify the effectiveness of the method. The method proposed by the present invention is validated by data collected in a tennessee-eastman (TEP) experiment. TEP is a small industrial process developed by eastman chemical company Downs and Vogel in 1993, the whole process consisting of five operating units including a chemical reactor, a condenser, a compressor, a vapor/liquid separator and a separator.
TEP contains eight ingredients: a, B, C, D, E, F, G and H, wherein gaseous species A, C, D and E and inert species B are reactants, G and H are reaction products, and F is a reaction byproduct.
Table 115 known faults (IDV)
Figure BDA0003369670880000161
Figure BDA0003369670880000171
The TEP co-generates 22 data sets for process monitoring and fault diagnosis, including 1 normal data and 8 quality-related fault training sets and 4 quality-independent fault training sets. In the training set, the normal data set contains 480 samples for establishing a non-linear model (i.e. the process of the steps S1 to S3 of the present invention), and the fault data set contains 480 fault samples for establishing a fault library; in the test set, each test data set contains 960 samples, the first 160 are normal samples and the last 800 are failure samples for experimental validation. Each input sample comprises 33 variables, and the test sample comprises 3 variables. The fault type IDV (1, 2, 6-8, 12, 13) is a quality-related fault data set (corresponding to a fault causing a system operation error), and the IDV (3, 4, 9, 15) is a quality-independent fault data set (corresponding to a disturbance not causing a system operation error).
The normal data set is substituted into the step S1 of the invention as a training sample, the parameters in the steps S2 and S3 are obtained through calculation, and then IDV (1-15) is used as the sensing data to be detected and is substituted into the step S4 of the invention to obtain the detection result shown in the table 2.
TABLE 2 quality-related Fault detection
Figure BDA0003369670880000172
Table 2 shows the detection condition of the quality-related fault in the fault IDV (8), and it can be seen from the detection rate that the present invention has good monitoring performance for the nonlinear chemical process, and the detection rate is greater than 90%. The specific detection result of the given fault IDV (14) is shown in fig. 2. In the IDV (14), the first 160 are normal samples, and the last 800 are fault samples, so that the normal samples are basically under the control limit, and the fault samples are basically effectively alarmed.
TABLE 3 quality independent Fault detection results
Figure BDA0003369670880000181
Table 3 shows the detection of class 4 quality independent faults. Among these faults, the process variable X has failed, but because it is quality independent, it does not cause a failure of a key performance indicator. Therefore, during the detection of the key performance indexes, the fault is not required to be alarmed, and if the fault is alarmed, the fault is a false alarm. As can be seen from Table 3, the false alarm rates of 4 types of mass-independent faults are all within 15%, wherein the faults 3, 4 and 9 are all less than 10%. The specific detection result given to the faulty IDV (4) is shown in fig. 3. In FIG. 3, T'2The statistic only has a small number of false alarms, and the quality-independent faults are effectively monitored in the quality-independent space of the SPE, so that the faults are identified, and the detection performance is good.

Claims (9)

1. A fault monitoring method for a weaponry system, comprising the steps of:
s1, collecting sensing data of the weapon equipment system, and carrying out linear standardization processing on the sensing data to obtain an input standard sensing data set and an output standard sensing data set;
s2, performing high-dimensional mapping and nonlinear processing on the input standard sensing data set and the output standard sensing data set to obtain a nonlinear input matrix, a nonlinear output matrix, an input parameter matrix and an output parameter matrix;
s3, calculating an input score matrix, an output score matrix, an input load matrix and an output load matrix according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix;
s4, collecting sensing data of the weapon equipment system to be detected to obtain the sensing data to be detected, and carrying out fault monitoring on the sensing data to be detected according to the input score matrix, the output score matrix, the input load matrix and the output load matrix.
2. The method for fault monitoring of a weaponry system of claim 1, wherein the type of data sensed in step S1 includes: voltage current data, pneumatic hydraulic data, pressure data, vibration data, temperature data, servo speed data, servo rotational speed data, and servo feedback voltage data, wherein the voltage current data, the pneumatic hydraulic data, the pressure data, the vibration data, and the temperature data are used as an input data set X, and the servo speed data, the servo rotational speed data, and the servo feedback voltage data are used as an output data set Y.
3. The method for monitoring faults of a weapons equipment system of claim 2, wherein the formula for linear normalization of the sensed data in step S1 is:
Figure FDA0003369670870000011
Figure FDA0003369670870000012
wherein ,
Figure FDA0003369670870000013
standard sensing data of j-th sampling sensing data for corresponding ith variable in input data set Xi，jSampling sensed data for the ith variable in the input data set X for the jth timei，jThe sensed data is sampled for the jth variable in the output data set Y,
Figure FDA0003369670870000021
the standard sensing data of j-th sampling sensing data of the corresponding ith variable in the output data set Y, n is the sampling times of each variable,
Figure FDA0003369670870000022
the constructed set is an input standard sensing dataset,
Figure FDA0003369670870000023
the constructed set is an output standard sensing dataset.
4. The fault monitoring method for a weaponry system of claim 3, wherein step S2 includes the substeps of:
s21, mapping the input standard sensing data set and the output standard sensing data set to a high-dimensional space respectively to obtain a high-dimensional input matrix phixAnd high dimensional output matrix phiy；
S22, inputting the matrix phi to the high dimensionxAnd high dimensional output matrix phiyPerforming nonlinear processing to obtain nonlinear input matrix
Figure FDA0003369670870000024
And a non-linear output matrix
Figure FDA0003369670870000025
S23, according to the nonlinear input matrix
Figure FDA0003369670870000026
And a non-linear output matrix
Figure FDA0003369670870000027
Calculating an input parameter matrix
Figure FDA0003369670870000028
And output parameter matrix
Figure FDA0003369670870000029
5. The method for fault monitoring of a weaponry system of claim 4 wherein the non-linear input matrix of step S22
Figure FDA00033696708700000210
And a non-linear output matrix
Figure FDA00033696708700000211
The calculation formula of (2) is as follows:
Figure FDA00033696708700000212
Figure FDA00033696708700000213
calculating an input parameter matrix in the step S23
Figure FDA00033696708700000214
And output parameter matrix
Figure FDA00033696708700000215
The formula of (1) is:
Figure FDA00033696708700000216
Figure FDA00033696708700000217
wherein 1n is
Figure FDA00033696708700000218
The vector of all 1 columns of the image,
Figure FDA00033696708700000219
is a real space.
6. The method for fault monitoring of a weapons equipment system of claim 5, wherein said step S3 includes the substeps of:
s31, constructing a nonlinear feature extraction target function according to the nonlinear input matrix, the nonlinear output matrix, the input parameter matrix and the output parameter matrix, and solving the nonlinear feature extraction target function by adopting Lagrange' S theorem to obtain an input-output parameter relation equation;
s32, calculating initial output latent variable component information according to the input and output parameter relation equation;
and S33, extracting an input score matrix, an output score matrix, an input load matrix and an output load matrix by a nonlinear regression iteration method according to the initial output latent variable component information.
7. The malfunction monitoring method for a weapons equipment system of claim 6, wherein the nonlinear feature extraction objective function of step S31 is:
Figure FDA0003369670870000031
the input and output parameter relation equation is as follows:
Figure FDA0003369670870000032
wherein w is a high-dimensional input matrix phixC is a high-dimensional output matrix phiyThe vector of the projection of (a) is,
Figure FDA0003369670870000033
in the case of a non-linear input matrix,
Figure FDA0003369670870000034
in the form of a non-linear output matrix,
Figure FDA0003369670870000035
in order to input the parameter matrix, the parameter matrix is,
Figure FDA0003369670870000036
and u is the initial output latent variable component information, and theta is the characteristic value corresponding to u after characteristic decomposition.
8. The method for fault monitoring of a weapons equipment system of claim 7, wherein said step S33 includes the substeps of:
s331, initializing the number of iterations i to 1, and outputting latent variable component information uiAssigning an initial value u;
s332, calculating input latent variable component information t of the ith iterationiAnd updating output latent variable component information ui；
Figure FDA0003369670870000037
Figure FDA0003369670870000038
wherein ,u′iFor updated output latent variable component information ui；
S333, from output latent variable component information u'iAnd input latent variable component information tiCalculating input load vector sumDeriving the load vector and applying to the non-linear input matrix
Figure FDA0003369670870000041
Non-linear output matrix
Figure FDA0003369670870000042
Input parameter matrix
Figure FDA0003369670870000043
And output parameter matrix
Figure FDA0003369670870000044
Updating is carried out;
Figure FDA0003369670870000045
Figure FDA0003369670870000046
Figure FDA0003369670870000047
Figure FDA0003369670870000048
Figure FDA0003369670870000049
Figure FDA00033696708700000410
wherein ,py，iTo output the load vector, px，iIs input negativeA vector number, E is an identity matrix,
Figure FDA00033696708700000411
for updated non-linear input matrix
Figure FDA00033696708700000412
For updated non-linear output matrix
Figure FDA00033696708700000413
For updated input parameter matrix
Figure FDA00033696708700000414
For updated output parameter matrix
Figure FDA00033696708700000415
S334, pair
Figure FDA00033696708700000416
In
Figure FDA00033696708700000417
Performing characteristic decomposition to obtain output latent variable component information ui+1；
S335, judging and outputting latent variable component information ui+1If yes, jumping to step S336, and if no, adding 1 to the iteration number i, and jumping to step S332;
s336, extracting A pieces of output latent variable component information uiA pieces of input latent variable component information tiA number of input load vectors px，iAnd A output load vectors py，iObtaining the input score matrix M ═ t (t)1，...，tA) And outputting a score matrix U ═ U1，...，uA) Input load matrix Px＝(px，1，...，px，A) And an output load matrix Py＝(py，1，...，py，A)。
9. The method for fault monitoring of a weapons equipment system of claim 8, wherein said step S4 includes the substeps of:
s41, calculating latent variables of single historical sensing data after dimension reduction projection:
Figure FDA00033696708700000418
wherein t is latent variable after dimension reduction projection of single historical sensing data, M is an input score matrix,
Figure FDA0003369670870000051
the method comprises the steps of performing Gaussian kernel functions on single historical sensing data and all historical sensing data, and performing standardization;
s42, according to the latent variable after dimension reduction projection is carried out on the single historical sensing data, establishing the quality-independent spatial statistic of the historical sensing data:
Q＝||φ(xi)-Pxt||2
wherein Q is the quality independent spatial statistic of the historical sensing data, phi (x)i) To be xiFunction for high dimensional mapping, xi＝{xi1，...，xij，...，xim}，|| ||2Performing two-norm operation;
s43, calculating a quality-related spatial control limit of the historical sensing data:
Figure FDA0003369670870000052
wherein ,
Figure FDA0003369670870000053
quality-dependent spatial control limits for historical sensory data, FA，n-AIs an F distribution;
s44, calculating the quality-independent space control limit of the historical sensing data according to the quality-independent space statistic of the historical sensing data:
Figure FDA0003369670870000054
wherein g is xi/2 mu, h is 2 mu2/ξ，Jth，SPEIs the quality-independent space control limit of the historical sensing data, g is a control limit coefficient, xi is the variance after all the historical sensing data construct a statistic Q, mu is the mean after all the historical sensing data construct the statistic Q,
Figure FDA0003369670870000055
is chi-square distribution, and h is the degree of freedom of chi-square distribution;
s45, collecting sensing data of the weapon equipment system to be tested to obtain the sensing data to be tested;
s46, mapping the to-be-detected sensing data after standardized processing to a high-dimensional space to obtain nonlinear to-be-detected sensing data;
s47, carrying out standardization processing on the Gaussian kernel function of the nonlinear sensing data to be measured to obtain the normalized Gaussian kernel function:
Figure FDA0003369670870000056
wherein ,
Figure FDA0003369670870000061
to normalize the processed Gaussian kernel function, knewThe method comprises the following steps that a Gaussian kernel function of nonlinear sensing data to be detected is adopted, and K is an input data set X and the Gaussian kernel function of the input data set X;
s48, projecting the nonlinear sensing data to be measured along the quality-related space according to the normalized Gaussian kernel function to obtain a latent variable tnew：
Figure FDA0003369670870000062
S49, calculating the quality-related spatial statistic and the quality-unrelated spatial statistic of the to-be-measured sensing data:
Figure FDA0003369670870000063
Figure FDA0003369670870000064
wherein ,T′2Is the quality-related spatial statistic of the sensed data to be measured, Q' is the quality-independent spatial statistic of the sensed data to be measured,
Figure FDA0003369670870000065
for the non-linear sensed data to be measured,
Figure FDA0003369670870000066
s50, judging quality related space statistic T 'of to-be-detected sensing data'2Whether or not to be greater than or equal to a quality-related spatial control limit of the historical sensory data
Figure FDA0003369670870000067
If yes, the existence of the weapon equipment system to be tested causes the fault of system operation error, namely, the existence of quality-related fault, if not, the step S51 is skipped to;
s51, judging whether the quality-independent space statistic Q' of the sensing data to be detected is more than or equal to the quality-independent space control limit J of the historical sensing datath，SPEIf the measured data of the weapon equipment system to be measured is normal data, the weapon equipment system to be measured operates normally.
CN202111394862.3A 2021-11-23 2021-11-23 Fault monitoring method for weapon equipment system Active CN114034332B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111394862.3A CN114034332B (en) 2021-11-23 2021-11-23 Fault monitoring method for weapon equipment system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111394862.3A CN114034332B (en) 2021-11-23 2021-11-23 Fault monitoring method for weapon equipment system

Publications (2)

Publication Number Publication Date
CN114034332A true CN114034332A (en) 2022-02-11
CN114034332B CN114034332B (en) 2023-09-22

Family

ID=80145222

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111394862.3A Active CN114034332B (en) 2021-11-23 2021-11-23 Fault monitoring method for weapon equipment system

Country Status (1)

Country Link
CN (1) CN114034332B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2431877A2 (en) * 2010-09-17 2012-03-21 Siemens Corporation System and method for modeling conditional dependence for anomaly detection in machine condition monitoring
CN107895224A (en) * 2017-10-30 2018-04-10 北京工业大学 A kind of MKECA fermentation process fault monitoring methods based on extension nuclear entropy load matrix
CN108664009A (en) * 2017-08-03 2018-10-16 湖州师范学院 Divided stages based on correlation analysis and fault detection method
CN110928262A (en) * 2019-12-17 2020-03-27 中国人民解放军火箭军工程大学 Quality-related fault online monitoring method for efficiently updating model in time-varying system
CN113420815A (en) * 2021-06-24 2021-09-21 江苏师范大学 Semi-supervised RSDAE nonlinear PLS intermittent process monitoring method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2431877A2 (en) * 2010-09-17 2012-03-21 Siemens Corporation System and method for modeling conditional dependence for anomaly detection in machine condition monitoring
CN108664009A (en) * 2017-08-03 2018-10-16 湖州师范学院 Divided stages based on correlation analysis and fault detection method
CN107895224A (en) * 2017-10-30 2018-04-10 北京工业大学 A kind of MKECA fermentation process fault monitoring methods based on extension nuclear entropy load matrix
CN110928262A (en) * 2019-12-17 2020-03-27 中国人民解放军火箭军工程大学 Quality-related fault online monitoring method for efficiently updating model in time-varying system
CN113420815A (en) * 2021-06-24 2021-09-21 江苏师范大学 Semi-supervised RSDAE nonlinear PLS intermittent process monitoring method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
CHUNHUI ZHAO 等: "The Multi-space Generalization of Total Projection to Latent Structures (MsT-PLS) and Its Application to Online Process Monitoring", 2013 10TH IEEE INTERNATIONAL CONFERENCE ON CONTROL AND AUTOMATION (ICCA), pages 1441 - 1446 *
孔祥玉 等: "基于偏最小二乘得分重构的质量相关故障检测", 控制理论与应用, vol. 37, no. 11, pages 2321 - 2332 *

Also Published As

Publication number Publication date
CN114034332B (en) 2023-09-22

Similar Documents

Publication Publication Date Title
CN109146246B (en) Fault detection method based on automatic encoder and Bayesian network
Amin et al. Robust process monitoring methodology for detection and diagnosis of unobservable faults
CN109145256B (en) Process monitoring method based on normative variable nonlinear principal component analysis
Ayodeji et al. Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction
CN111639304B (en) CSTR fault positioning method based on Xgboost regression model
CN111340110A (en) Fault early warning method based on industrial process running state trend analysis
Kopbayev et al. Fault detection and diagnosis to enhance safety in digitalized process system
CN113703422B (en) Gas turbine pneumatic actuator fault diagnosis method based on feature analysis processing
CN110942258B (en) Performance-driven industrial process anomaly monitoring method
Luo et al. Key-performance-indicators-related fault subspace extraction for the reconstruction-based fault diagnosis
CN114297921A (en) AM-TCN-based fault diagnosis method
CN117313015A (en) Time sequence abnormality detection method and system based on time sequence and multiple variables
CN113009823A (en) Method and apparatus for controlling a machine
Fei et al. Online process monitoring for complex systems with dynamic weighted principal component analysis
Dong et al. Dynamic-inner canonical correlation analysis based process monitoring
CN114034332A (en) Fault monitoring method for weapon equipment system
Yang et al. A semi-supervised feature contrast convolutional neural network for processes fault diagnosis
Xie et al. Fault detection based on probabilistic kernel partial least square regression for industrial processes
CN109523195B (en) Sample multi-modal classification-based fault detection method for hierarchical non-Gaussian algorithm
CN114383648A (en) Temperature instrument fault diagnosis method and device
CN113076211B (en) Quality-related fault diagnosis and false alarm feedback method based on fault reconstruction
CN114112374A (en) Fault diagnosis method of weapon equipment system based on orthogonal sub-block projection
Ma et al. Process monitoring of the pneumatic control valve using canonical variate analysis
Huang et al. A fault detection method based on stacking the SAE-SRBM for nonstationary and stationary hybrid processes
CN110928263B (en) Fault detection method and system for complex process considering dynamic relationship in advance

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