CN113359679A - Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics - Google Patents
Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics Download PDFInfo
- Publication number
- CN113359679A CN113359679A CN202110715891.9A CN202110715891A CN113359679A CN 113359679 A CN113359679 A CN 113359679A CN 202110715891 A CN202110715891 A CN 202110715891A CN 113359679 A CN113359679 A CN 113359679A
- Authority
- CN
- China
- Prior art keywords
- fault
- data
- historical
- amplitude
- dynamic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 27
- 238000003745 diagnosis Methods 0.000 title claims abstract description 22
- 230000002159 abnormal effect Effects 0.000 claims abstract description 37
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000000354 decomposition reaction Methods 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 10
- 238000000513 principal component analysis Methods 0.000 claims description 9
- 239000000126 substance Substances 0.000 claims description 7
- 230000003068 static effect Effects 0.000 claims description 2
- 101150096839 Fcmr gene Proteins 0.000 claims 1
- 238000005259 measurement Methods 0.000 description 9
- 238000000926 separation method Methods 0.000 description 8
- 238000012360 testing method Methods 0.000 description 6
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 230000002411 adverse Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0262—Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
-
- 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
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24065—Real time diagnostics
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to an industrial process fault diagnosis method based on reconstructed amplitude trend characteristics, which particularly comprises the following steps: detecting real-time data of the industrial process based on a pre-established latent structure model, and determining whether the detected data is abnormal; the latent structure model is established in advance by adopting historical normal data; when abnormal data exist in the detected data, acquiring a fault amplitude of the abnormal data; acquiring the similarity of each fault amplitude in the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults based on a dynamic time warping algorithm; and determining the fault type of the abnormal data according to the comparison result of the similarity and the specified threshold. According to the fault diagnosis method, the fault data are reconstructed, the fault amplitude is estimated, and then the similarity of the fault amplitude sequence is calculated by utilizing dynamic time distortion, so that faults with similar fault directions and different types can be separated.
Description
Technical Field
The invention relates to an industrial production process control technology, in particular to an industrial process fault diagnosis method based on reconstructed amplitude trend characteristics.
Background
Fault isolation is one of the most important tasks in the field of industrial fault diagnosis. And the fault separation is to judge the fault type of the equipment by using the process related variables. In the industrial process, the production efficiency can be reduced, the machine can be stopped and even the personnel can be injured due to the failure of equipment during operation. When a fault occurs, how to correctly separate the fault is important to take proper measures to quickly restore the process to normal. Therefore, fault separation is of great significance in ensuring safe and smooth operation of the industrial process.
The traditional fault separation method mainly utilizes fault directions to separate faults, and does not consider the problem of separation among faults with similar fault directions but different types.
Therefore, the separation of faults with similar fault directions but different types becomes a hot spot of research in the industry.
Disclosure of Invention
Technical problem to be solved
In view of the above disadvantages and shortcomings of the prior art, the present invention provides a fault diagnosis method based on reconstructed amplitude trend characteristics, which is used to separate faults with similar fault directions but different types.
The reconstruction trend characteristic refers to that a fault amplitude sequence is obtained through reconstruction, and fault diagnosis is performed by taking the trend of the fault amplitude sequence as a characteristic, so that the reconstruction trend characteristic is called.
(II) technical scheme
In order to achieve the purpose, the invention adopts the main technical scheme that:
an industrial process fault diagnosis method based on reconstructed amplitude trend characteristics comprises the following steps:
s1, detecting real-time data of the industrial process based on the pre-established latent structure model, and confirming whether the detected data have abnormal data; the latent structure model is established in advance by adopting historical normal data;
s2, acquiring the fault amplitude of the abnormal data when the abnormal data exists in the detected data;
s3, acquiring the similarity of each fault amplitude in the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults based on a dynamic time warping algorithm;
and S4, determining the fault type of the abnormal data according to the comparison result of the similarity and the specified threshold.
(III) advantageous effects
According to the fault diagnosis method provided by the invention, the amplitude of the fault is estimated by reconstruction for the faults with similar fault directions and different types, then a new fault similarity is defined by using the fault amplitude characteristic, and the fault amplitude sequence is obtained by reconstruction.
Different from the traditional contribution graph algorithm, the fault diagnosis method provided by the application firstly establishes a dynamic latent structure model by using historical normal data, and obtains a historical fault direction by using singular value decomposition on the historical fault data; the historical fault data is then projected along the historical fault direction to estimate the fault magnitude. For real-time data, firstly, judging whether the real-time data has a fault by using a control limit obtained by normal data, and if the real-time data does not have the fault, continuing monitoring; otherwise, executing the step of fault diagnosis. Firstly, the fault direction is not extracted by using singular value decomposition, but projected along all the historical fault directions obtained before respectively, and a fault amplitude sequence is estimated. And carrying out similarity measurement on the obtained fault amplitude sequence and amplitude sequences of all historical fault data to obtain corresponding similarity. If all the similarity degrees are lower than the set threshold value, the fault is regarded as a novel fault and added into a historical fault set; otherwise, the type of the historical fault corresponding to the amplitude sequence with the maximum similarity is the fault type of the test data.
Because the fault data is reconstructed and the fault amplitude is estimated, and then the similarity measurement is carried out on the fault amplitude sequence by using dynamic time warping, faults with similar fault directions but different types can be separated.
Drawings
FIG. 1 is a flow chart of a method for diagnosing faults in an industrial process based on reconstructed amplitude trend characteristics according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for diagnosing faults in an industrial process based on reconstructed amplitude trend characteristics according to another embodiment of the present invention;
fig. 3 is a schematic diagram of the failure directions of two types of failures according to an embodiment of the present invention.
FIG. 4 is a diagram illustrating dynamic time warping of amplitude sequences of two types of historical faults and amplitude sequences of two types of test faults according to an embodiment of the present invention.
Detailed Description
For the purpose of better explaining the present invention and to facilitate understanding, the present invention will be described in detail by way of specific embodiments with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
When the equipment is in operation in an industrial process, the failure can cause the reduction of production efficiency, the shutdown and even the casualties. After the fault occurs, how to correctly separate the fault is important to adopt proper measures to quickly restore the process to normal. Therefore, fault separation is of great significance in ensuring safe and smooth operation of the industrial process. However, the existing fault separation method mainly utilizes the fault direction to separate faults, and does not consider the problem of separation between faults with similar fault directions but different types.
Based on the method, the fault data are reconstructed, the fault amplitude is estimated, and then the fault amplitude sequence is subjected to similarity measurement by utilizing dynamic time distortion, so that faults with similar fault directions and different types can be separated.
Example one
As shown in fig. 1 and 2, the present embodiment provides a method for diagnosing a fault of an industrial process based on reconstructed amplitude trend characteristics, the method comprising the steps of:
s1, detecting real-time data of the industrial process based on the pre-established latent structure model, and confirming whether the detected data have abnormal data; the latent structure model is established in advance by adopting historical normal data;
s2, acquiring the fault amplitude of the abnormal data when the abnormal data exists in the detected data;
s3, acquiring the similarity of each fault amplitude in the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults based on a dynamic time warping algorithm;
and S4, determining the fault type of the abnormal data according to the comparison result of the similarity and the specified threshold.
In this embodiment, similarity measurement is performed on the obtained fault amplitude sequence and amplitude sequences of all historical fault data to obtain corresponding similarity. If all the similarity degrees are lower than the set threshold value, the fault is regarded as a novel fault and added into a historical fault set; otherwise, the type of the historical fault corresponding to the amplitude sequence with the maximum similarity is the fault type of the test data. Because the fault data is reconstructed and the fault amplitude is estimated, and then the similarity measurement is carried out on the fault amplitude sequence by using dynamic time warping, faults with similar fault directions but different types can be separated.
Further, in order to obtain the amplitude sequence of the historical fault data, before the step S1, the method for diagnosing the fault of the industrial process according to the embodiment further includes:
s0, acquiring normal historical data in the industrial process within a first preset time period, establishing a latent structure model for the normal historical data by utilizing dynamic internal principal component analysis, and calculating a corresponding comprehensive index control limit.
The comprehensive index control limit is used for measuring whether the test sample is abnormal or not, subsequent real-time data can be monitored and used for judging whether the real-time data is a fault or not, and if the real-time data is not the fault, the monitoring is continued.
Further, before the step S3, the method for diagnosing faults in an industrial process according to this embodiment further includes:
s3a, acquiring abnormal historical fault data in the industrial process in a second preset time period, decomposing each historical fault data by using singular values, and extracting a historical fault direction; and
reconstructing each historical fault data along the historical fault direction to which the historical fault data belongs, and estimating the fault amplitude of the historical fault data;
and forming a fault amplitude sequence by using the fault amplitudes of all the historical fault data.
By fault direction is understood that when a fault occurs, each variable has a value, and the values form a vector in a high-dimensional space, so that the fault direction is also called as a fault direction vector.
The fault amplitude sequence is composed of the fault amplitudes estimated from each historical fault data according to time.
Wherein, the step S3 specifically includes the following steps:
s31, calculating a dynamic time warping distance between the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical fault;
and S32, calculating the dynamic time distortion similarity between the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical fault.
And comparing with a threshold value after the dynamic time warping similarity is obtained. The threshold value can be set according to experience or task goals, and can also be preset according to prior knowledge. When the similarity is larger than a threshold value, taking the historical fault type in the similarity as the fault type of the current fault data; if the similarity is less than the threshold, the fault type of the current fault data is defined, that is, step S4.
When similarity calculation is carried out, the method aims at a fault amplitude sequence which is obtained by dynamic intrinsic principal component analysis and reconstruction-based method estimation and is not a traditional original variable curve of an object, so that large normal fluctuation under normal working conditions can be effectively removed, and adverse effects of the method on fault diagnosis are avoided.
In the embodiment, the method firstly establishes a dynamic latent structure model by using historical normal data, and then projects the historical fault data along the historical fault direction, so as to estimate the fault amplitude. For real-time data, firstly, a dynamic latent structure model obtained through normal data is used for calculating a comprehensive index, whether the real-time data is in fault or not is judged according to whether the comprehensive index exceeds a comprehensive index control limit or not, and if the real-time data is not in fault, monitoring is continued; otherwise, executing a fault diagnosis step, projecting the fault diagnosis step along all the historical fault directions obtained before respectively, and estimating a fault amplitude sequence. And carrying out similarity measurement on the obtained fault amplitude sequence and amplitude sequences of all historical fault data to obtain corresponding similarity. If all the similarity degrees are lower than the set threshold value, the fault is regarded as a novel fault and added into a historical fault set; otherwise, the type of the historical fault corresponding to the amplitude sequence with the maximum similarity is the fault type of the test data. Faults of similar direction but different types can be accurately separated.
Example two
In order to better understand the above model and the updating process of the model parameters, the present invention is described in detail again by another embodiment.
As shown in fig. 2, the method for diagnosing a fault in an industrial process based on a reconstructed amplitude trend feature provided in this embodiment includes the following specific steps:
step 1: modeling normal historical data; and establishing a latent structure model for normal historical data by utilizing dynamic internal pivot analysis and calculating related control limits. The control limit is a comprehensive index control limit.
The specific steps of the step 1 are as follows:
for a process data matrix with m-dimensional variables, N samples
Establishing a latent structure model by utilizing dynamic internal pivot analysis, wherein the samples are historical normal data:
wherein t isjIs a dynamic latent variable at the moment j and s is a dynamic order and can pass cross checkConfirming the certificate; beta is ai(i ═ 1, 2.., s) are coefficients of an autoregressive model; t is tj-iIs a dynamic latent variable i times before the current j time, vjThe model error of the dynamic latent variable at the current moment is taken as the model error of the dynamic latent variable at the current moment;
p is a dynamic load matrix and is,
the predicted value of the dynamic latent variable is obtained;
is the model residual of the current time sample, PrIs ejLoad matrix, t, obtained by principal component analysis modelingr,jIs ejLatent variable of er,jIs ejAnd (4) performing principal component analysis modeling on the static residual error part. The number of principal elements is determined by the principle of cumulative variance. Each control limit is then calculated, and for ease of description only the composite index control limit is considered here
Wherein the content of the first and second substances,
wherein S is a covariance matrix of the sample matrix X;
wherein the content of the first and second substances,
and
are respectively paired with ejSPE index control limit and T obtained by principal component analysis modeling2An index control limit;
Step 2: extracting historical fault directions; for each historical fault data in the historical fault data set, the historical fault direction is extracted for its residual using singular value decomposition, as shown in fig. 3.
The method comprises the following specific steps:
for historical failure data set
Each of the historical failure data X inl(L ═ 1, 2.. times, L), for which the residuals are left
Singular value decomposition is carried out:
the fault direction matrix is then:
wherein, UhIs to
Matrices formed by left singular vectors when singular value decomposition is performed, DhIs to
When singular value decomposition is carried out, a diagonal matrix formed by singular values is formed;
is to
And (4) performing matrix composed of right singular vectors when singular value decomposition is performed.
Since the residual matrix and the fault direction matrix have the same column space, the residual is corrected
The matrix formed by the left singular vectors obtained after singular value decomposition can be regarded as a fault direction matrix.
In this embodiment, only a single fault diagnosis problem is considered, and therefore, the fault direction matrix
Can be degraded into fault direction vectors
The single fault, that is, the single-dimensional fault, means that only one direction has a fault, and the corresponding direction is the multi-dimensional fault, which means that a plurality of directions have faults.
And step 3: and reconstructing historical faults and estimating fault amplitude values corresponding to the historical faults.
The reconstruction of the historical faults is to reconstruct the data X of each historical faultlResidual error e ofjReconstructing along the corresponding historical fault direction extracted in step 2, and estimating the corresponding fault amplitude, specifically as follows:
each historical fault data in the historical fault data set is arranged along the corresponding fault direction
And (3) reconstruction:
wherein
For the residual of the reconstructed process data,
the magnitude of the fault estimated from the reconstruction.
It should be noted that, in the industrial process fault diagnosis method provided by the embodiment, the reconstructed object is not the original sample, i.e. not the historical data or the detected real-time data, but the residual e of the original samplejSo as to remove large normal fluctuation under normal working conditions and avoid adverse effects on fault diagnosis.
The residual error in this embodiment refers to sample data, including historical fault data and a dynamic latent structure model residual error of real-time detection fault data.
The reconstruction aims at the comprehensive index of the reconstructed sample
The value is minimized, and the reconstructed composite index can be expressed as:
namely:
to seek
Is a minimum value of (2), will synthesize the index
For fault amplitude
Calculating a partial derivative:
let the above formula equal to 0, solve the fault amplitude
And 4, step 4: detecting on line; applying the model in the step 1 to the detection of real-time data, calculating related indexes, and if the indexes exceed the corresponding comprehensive index control limit in the step 1, determining that the data is abnormal, specifically as follows:
for real-time sampled data xjIn this embodiment, for the sake of clarity, only the comprehensive indexes are taken as an example for detailed description:
if it is not
Then the data x is consideredjIs the exception data.
Wherein the content of the first and second substances,
is the control limit of the comprehensive index.
And 5: estimating the fault amplitude of the real-time abnormal data, specifically reconstructing the residual error of the abnormal data in the step 4 along each historical fault direction to obtain the corresponding fault amplitude
Real-time abnormal data xjIs along the residual errorFrom the respective fault directions extracted in step 2
And (3) reconstruction:
step 6: as shown in fig. 4, the similarity is calculated; and 5, calculating the similarity of the fault amplitude sequence reconstructed by the residual error of the real-time abnormal data obtained in the step 5 along the historical fault direction and the fault amplitude sequence of each historical fault by using dynamic time warping.
Dynamic time warping is to better match and map the time sequence morphology by warping the time axis for similarity measurement, which allows point-to-multipoint nonlinear alignment to find the best matching way of two time sequences, and its essence is to find a path in a cost matrix by using dynamic programming to determine the lowest matching cost. The cost matrix records the distance between any two points in the two time series. The task of dynamic time warping is to find a curved path from the bottom left to the top right in the cost matrix such that the distance between the two time series is minimized. The dynamic time warping algorithm comprises the following specific steps:
step 6-1, calculating the Euclidean distance between any points of two fault amplitude sequences through the following formula, wherein one of the two fault amplitude sequences is a fault amplitude sequence reconstructed by the residual error of the real-time abnormal data along each historical fault direction, the other two fault amplitude sequences are fault amplitude sequences of historical faults,
and order
Wherein i is more than or equal to 1 and less than or equal to k0，1≤i≤kl；
Wherein f isc＝{fc(1),fc(2),fc(3),…,fc(k0) Is the fault to be diagnosed
The sequence of amplitudes of the first and second signals,
representing the ith historical failure
L, L is a historical fault set
The size of (2).
Step 6-2, defining the distortion path of the two fault amplitude sequence as W ═ W1,w2,…,wkWhere max (k)0,kl)≤k≤k0+kl，
The purpose of dynamic time warping is to find a time from M (1,1) to M (k)0,kl) So that
A monotonically increasing path to the minimum is taken.
The optimal path is found according to the following formula:
and 6-3, calculating the dynamic time distortion distance between the two fault amplitude sequences according to the following formula:
6-4, obtaining the distance between the fault amplitude sequence to be diagnosed and all historical fault amplitude sequences
Then, the dynamic time warping similarity of the two is calculated according to the following formula:
wherein the content of the first and second substances,
is to diagnose the fault
Amplitude sequence f ofcAnd the first history fault
Amplitude sequence of
A dynamic time warping distance between; drefIs an empirically set dynamic time warp distance reference value.
In the step 6, the sequence similarity calculation based on dynamic time warping is not a traditional original variable curve, but a fault amplitude sequence estimated by using dynamic intrinsic principal component analysis and a reconstruction-based method. Thus, the method is also beneficial to removing large normal fluctuation under normal working conditions and avoiding adverse effects on fault diagnosis.
And 7: diagnosing faults; and if all the obtained similarity degrees are less than a threshold value, the fault is considered to be a new fault, a historical fault set is added, and the step 2 is returned. Otherwise, the historical fault type with the maximum similarity is the fault type of the real-time fault data.
If all the similarity degrees Sc,lAre all less than a threshold th1If the fault is a new fault, adding the historical fault set
And returning to the step 2. Otherwise, the historical fault type with the maximum similarity
I.e. real-time fault data xjThe type of failure of (2).
The threshold value may be set according to experience or a task target, or may be preset according to a priori knowledge.
The fault diagnosis method provided by the embodiment is different from the traditional contribution graph algorithm, the algorithm firstly utilizes historical normal data to establish a dynamic latent structure model, and utilizes singular value decomposition on the historical fault data to obtain a historical fault direction; the historical fault data is then projected along the historical fault direction to estimate the fault magnitude. For real-time data, firstly, judging whether the real-time data has a fault by using a control limit obtained by normal data, and if the real-time data does not have the fault, continuing monitoring; otherwise, executing the step of fault diagnosis. Firstly, the fault direction is not extracted by using singular value decomposition, but projected along all the historical fault directions obtained before respectively, and a fault amplitude sequence is estimated. And carrying out similarity measurement on the obtained fault amplitude sequence and amplitude sequences of all historical fault data to obtain corresponding similarity. If all the similarity degrees are lower than the set threshold value, the fault is regarded as a novel fault and added into a historical fault set; otherwise, the type of the historical fault corresponding to the amplitude sequence with the maximum similarity is the fault type of the test data.
Because the fault data is reconstructed and the fault amplitude is estimated, and then the similarity measurement is carried out on the fault amplitude sequence by using dynamic time warping, faults with similar fault directions but different types can be separated.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, the claims should be construed to include preferred embodiments and all changes and modifications that fall within the scope of the invention.
It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include such modifications and variations.
Claims (8)
1. An industrial process fault diagnosis method based on reconstructed amplitude trend characteristics is characterized by comprising the following steps:
s1, detecting real-time data of the industrial process based on the pre-established latent structure model, and confirming whether the detected data have abnormal data; the latent structure model is established in advance by adopting historical normal data;
s2, acquiring the fault amplitude of the abnormal data when the abnormal data exists in the detected data;
s3, acquiring the similarity of each fault amplitude in the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults based on a dynamic time warping algorithm;
and S4, determining the fault type of the abnormal data according to the comparison result of the similarity and the specified threshold.
2. The method according to claim 1, wherein before the S1, further comprising:
s0, acquiring normal historical data in the industrial process within a first preset time period, establishing a latent structure model for the normal historical data by utilizing dynamic internal principal component analysis, and calculating a corresponding comprehensive index control limit.
3. The method according to claim 1 or 2, wherein the S3 is preceded by:
s3a, acquiring abnormal historical fault data in the industrial process in a second preset time period, decomposing each historical fault data by using singular values, and extracting a historical fault direction; and
reconstructing each historical fault data along the historical fault direction to which the historical fault data belongs, and estimating the fault amplitude of the historical fault data;
and forming a fault amplitude sequence by using the fault amplitudes of all the historical fault data.
4. The method of claim 2,
the latent structure model is as follows:
wherein, tjIs the dynamic latent variable at time j, s is the dynamic order, betai(i ═ 1, 2.. times, s) are coefficients of an autoregressive model, xjIs the sampled data at time j, tj-iIs a dynamic latent variable i times before the current j time, vjThe model error of the dynamic latent variable at the current moment is taken as the model error of the dynamic latent variable at the current moment;
p is a dynamic load matrix and is,
as a predictor of the dynamic latent variable, ejIs the model residual of the current time sample, PrIs ejLoad matrix, t, obtained by principal component analysis modelingr，jIs ejLatent variable of er，jIs a residual error ejAnd (4) performing principal component analysis modeling on the static residual error part.
5. The method of claim 3, wherein S3a comprises:
the fault amplitude of the historical fault data is obtained according to the following formula:
wherein phiePositive definite matrix for projecting original variables as statistical indexes, ejIn order to be a dynamic latent structure model residual,
the historical fault direction vector is used as the fault direction vector if the fault is a single fault
From a fault direction matrix
Degenerated, said fault direction matrix
For decomposing each said historical fault data X by using singular valuesl(L ═ 1, 2.., L), and obtained according to the following formula:
wherein the content of the first and second substances,
for historical fault data sets
Each of the historical failure data X inlResidual error of (1, 2.., L), UhIs to
Matrix of left singular vectors in singular value decomposition, DhIs to
So that the diagonal matrix composed of singular values in singular value decomposition,
is to
And (4) performing matrix composed of right singular vectors when singular value decomposition is performed.
6. The method according to claim 1, wherein the S1 includes:
for real-time sampled data xjAnd calculating a comprehensive index by adopting the latent structure model:
if it is not
Confirm the data xjIs abnormal data;
the S2 includes:
following the residuals of the anomaly data along each fault direction extracted in step 2
And (3) reconstruction:
wherein the content of the first and second substances,
the fault amplitude of the abnormal data.
7. The method according to claim 1, wherein the S3 includes:
obtaining the similarity of each fault amplitude in the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults based on a dynamic time warping algorithm
S31, calculating the dynamic time warping distance between the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults according to the following formula:
wherein the content of the first and second substances,
is to diagnose the fault
Amplitude sequence f ofcAnd the first history fault
Amplitude sequence of
A dynamic time warping distance between; fc ═ fc(1)，fc(2)，ft(3)，...，fc(k0) Is the fault to be diagnosed
The sequence of amplitudes of the first and second signals,
representing the ith historical failure
L, L is a historical fault set
The size of (d);
s32, calculating the dynamic time distortion similarity between the fault amplitude sequence of the abnormal data and the fault amplitude sequence of the historical faults according to the following formula:
wherein D isrefIs an empirically set dynamic time warp distance reference value.
8. The method according to claim 1, wherein the S4 includes:
when the similarity is larger than a threshold value, taking the historical fault type in the similarity as the fault type of the current fault data; and if the similarity is smaller than the threshold value, defining the fault type of the current abnormal data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110715891.9A CN113359679A (en) | 2021-06-24 | 2021-06-24 | Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110715891.9A CN113359679A (en) | 2021-06-24 | 2021-06-24 | Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113359679A true CN113359679A (en) | 2021-09-07 |
Family
ID=77536703
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110715891.9A Pending CN113359679A (en) | 2021-06-24 | 2021-06-24 | Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113359679A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116027724A (en) * | 2022-09-23 | 2023-04-28 | 河北东来工程技术服务有限公司 | Ship equipment risk monitoring method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102110187A (en) * | 2009-12-28 | 2011-06-29 | 清华大学 | Method and system for diagnosing mixed failure based on PCA and artificial immune system |
| CN109799808A (en) * | 2019-01-25 | 2019-05-24 | 东北大学 | A kind of dynamic process failure prediction method based on reconfiguration technique |
| CN110083860A (en) * | 2019-03-13 | 2019-08-02 | 东北大学 | A kind of industrial method for diagnosing faults based on correlated variables selection |
-
2021
- 2021-06-24 CN CN202110715891.9A patent/CN113359679A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102110187A (en) * | 2009-12-28 | 2011-06-29 | 清华大学 | Method and system for diagnosing mixed failure based on PCA and artificial immune system |
| CN109799808A (en) * | 2019-01-25 | 2019-05-24 | 东北大学 | A kind of dynamic process failure prediction method based on reconfiguration technique |
| CN110083860A (en) * | 2019-03-13 | 2019-08-02 | 东北大学 | A kind of industrial method for diagnosing faults based on correlated variables selection |
Non-Patent Citations (5)
| Title |
|---|
| GANG LI 等: "Dynamic time warping based causality analysis for root-cause diagnosis of nonstationary fault processes", IFAC-PAPERSONLINE, vol. 48, no. 8, 25 September 2015 (2015-09-25), pages 1288 - 1293 * |
| LIJIE ZHAO 等: "Adaptive moving window MPCA for online batch monitoring", 2004 5TH ASIAN CONTROL CONFERENCE, 9 May 2005 (2005-05-09) * |
| XIAOGANG DENG 等: "Fault Diagnosis Method Based on Independent Component Analysis and Dynamic Time Warping", 2006 6TH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION, 23 October 2006 (2006-10-23) * |
| YUNPENG CAO 等: "Start-up Fault Detection of Gas Turbine Based on Dynamic Time Warping", 2019 INTERNATIONAL CONFERENCE ON SENSING, DIAGNOSTICS, PROGNOSTICS, AND CONTROL, 17 August 2020 (2020-08-17) * |
| 刘强 等: "数据驱动的工业过程运行监控与自优化研究展望", 自动化学报, 22 November 2018 (2018-11-22) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116027724A (en) * | 2022-09-23 | 2023-04-28 | 河北东来工程技术服务有限公司 | Ship equipment risk monitoring method and system |
| CN116027724B (en) * | 2022-09-23 | 2024-01-12 | 河北东来工程技术服务有限公司 | Ship equipment risk monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
| CN107632592B (en) | Nonlinear time-varying process fault monitoring method based on efficient recursion kernel principal component analysis | |
| CN110298455B (en) | Mechanical equipment fault intelligent early warning method based on multivariate estimation prediction | |
| CN111275288A (en) | XGboost-based multi-dimensional data anomaly detection method and device | |
| CN109270907B (en) | Process monitoring and fault diagnosis method based on hierarchical probability density decomposition | |
| CN110083860B (en) | Industrial fault diagnosis method based on relevant variable selection | |
| CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
| CN111368428B (en) | Sensor precision degradation fault detection method based on monitoring second-order statistics | |
| CN115800272B (en) | Power grid fault analysis method, system, terminal and medium based on topology identification | |
| CN113239132A (en) | Online out-of-tolerance identification method for voltage transformer | |
| CN110990788A (en) | Bearing residual life prediction method based on ternary wiener process | |
| CN113359679A (en) | Industrial process fault diagnosis method based on reconstructed amplitude trend characteristics | |
| CN116431966A (en) | Reactor core temperature anomaly detection method of incremental characteristic decoupling self-encoder | |
| CN110209145A (en) | One kind being based on the approximate carbon dioxide absorption tower method for diagnosing faults of nuclear matrix | |
| Zhao et al. | Comprehensive subspace decomposition and isolation of principal reconstruction directions for online fault diagnosis | |
| CN117350377A (en) | Knowledge graph driving-based equipment fault diagnosis method and device | |
| CN113159088A (en) | Fault monitoring and diagnosis method based on multi-feature fusion and width learning | |
| CN115047853B (en) | Minor fault detection method based on recursion standard variable residual error and kernel principal component analysis | |
| Uchida et al. | Process fault diagnosis method based on MSPC and LiNGAM and its application to Tennessee Eastman process | |
| CN115327436A (en) | Power distribution network short-circuit fault diagnosis method based on composite statistics | |
| CN116048036A (en) | Fault detection method based on dynamic non-stationary projection structure | |
| CN115792636A (en) | Battery pack tiny fault quantitative diagnosis method and system based on maximum voltage value | |
| CN113435106B (en) | Transition mode operation fault detection method and system | |
| KR102486463B1 (en) | Method and Apparatus for Real Time Fault Detection Using Time series data According to Degradation | |
| CN114200914A (en) | MW-OCCA-based quality-related early fault detection method |
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 |