CN105955241B - A kind of quality fault localization method based on joint data-driven production process - Google Patents
A kind of quality fault localization method based on joint data-driven production process Download PDFInfo
- Publication number
- CN105955241B CN105955241B CN201610391112.3A CN201610391112A CN105955241B CN 105955241 B CN105955241 B CN 105955241B CN 201610391112 A CN201610391112 A CN 201610391112A CN 105955241 B CN105955241 B CN 105955241B
- Authority
- CN
- China
- Prior art keywords
- quality
- fault
- data
- cause
- modal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- 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
Abstract
The present invention provides a kind of quality fault localization method based on joint data-driven production process, and this method includes:Extract quality cause and effect topology graph model;Establish the multi-modal monitoring model of joint data-driven;The Performance Evaluation index of quality fault diagnosis is established based on contribution rate and procedural knowledge;According to the multi-modal monitoring model, quality fault propagation path is identified, the Performance Evaluation index alignment quality failure diagnosed according to the quality fault.The present invention is on the basis of topological diagram feature extraction, the process monitoring of multivariate statistics data-driven and machine learning, propose the fault diagnosis for the joint data-driven for being suitable for quality monitoring, to provide new approach based on the diagnosis of the production process quality fault of data and knowledge, the insoluble quality fault propagation path identification of traditional statistical process monitoring and fault-location problem are compensated for, accurately and efficiently quality fault positioning and diagnosis are realized with the data-driven of combining of " quantitative and qualitative is quantitative " of knowledge based on data.
Description
Technical field
The invention belongs to the control of production process and monitoring technical fields, and in particular to one kind is given birth to based on joint data-driven
The quality fault localization method of production process.
Background technology
Batch production is the division of labor refinement to modern production process and procedure process, is widely used in machinery, five
The industries such as gold, plastics, automobile fitting.In recent years, to adapt to demand of the market to multi items, more specifications, high quality functional product,
Industrial process of having a rest just develops towards efficient, large-scale and integrated direction, and as production-scale expansion and complexity increase,
The safety and reliability of production process is required also higher and higher.Often variable and control loop are many for modern complexity batch process
More and interrelated, a node failure will directly influence product quality and productivity effect, or even cause production process
Paralysis, will cause a serious accident for the manufacturing enterprises such as steel, coloured, chemical industry if failure cannot be diagnosed and be excluded in time.For
Ensure the safety of production process, the stability of product quality, complicated batch process is monitored on-line, accurately carries out event
Barrier diagnosis, and debug in time, ensure that end product quality meets the requirements and has become one of current process control field
Important research direction.
For example, the modern hot strip rolling based on joint data-driven is one by the high-quality, high of order Flexible Production
The full-automatic production operation line of effect, about 3,500,000 tons of strips of typical 1700mm hot strip rollings year output, mill speed are reachable
20m/s, finished product thickness 0.8~12.7mm of range, 700~1550mm of width range, can cover hundreds of steel grades.Finished strip
Surface quality, internal flaw, plate shape, thickness, width and structure property directly affect deep processing and the material property of strip, also directly
Connecing influences the economic benefit of enterprise.Hot strip rolling whole process has nearly 15000 process variables, control loop number to have nearly 300
Control loop, nearly half process variable directly or indirectly influence belt steel product quality.These process variables and control loop are mutual
It influences and is associated with, occur being difficult that accurately and timely judgement is related when product quality (the especially quality such as plate shape, structure property) fluctuation
The reason of failure, causes certain enterprises often to stop production repair (the often random all fronts of milli because product quality user returns goods
It safeguards).
The basic reason for generating above-mentioned puzzlement is that this complicated batch process of hot strip rolling has inherent move
State is non-linear, the time-varying characteristics of multimode step response caused by close coupling, multiple batches of multi-state between variable and circuit, process, random
The features such as uncertainty that noise generates, cause quality fault reason is various, failure evolution process is complicated, failure specific location and
Change direction is uncertain, fault coverage is wide in range, there are juxtapositions etc. for failure and reason, and traditional course monitoring method is in mistake
It is excessively coarse in the description of journey, cannot the priori of fully mining process be applied to monitoring to quality fault and therefore answering
There is significant limitation in, in time, accurately the quality fault of production process cannot effectively monitor and judge.
Invention content
The purpose of the embodiment of the present invention is to provide a kind of quality fault positioning side driving production process based on data aggregate
Method to the quality fault of production process effectively monitor and judge in time, accurately.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of matter driving production process based on data aggregate
Fault Locating Method is measured, described method includes following steps:
Extract quality cause and effect topology graph model;
Establish the multi-modal monitoring model of joint data-driven;
The Performance Evaluation index of quality fault diagnosis is established based on contribution rate and procedural knowledge;
According to the multi-modal monitoring model, quality fault propagation path is identified, the property diagnosed according to the quality fault
It can evaluation index alignment quality failure.
In said program, the extraction quality cause and effect topology graph model, the further process to be driven based on data aggregate
Knowledge extracts quality cause and effect topology graph model with historical data.
In said program, the extraction quality cause and effect topology graph model specifically comprises the following steps:
Step 101, quality cause and effect topological diagram correlation of variables is analyzed;
Step 102, between design variable correlation metric threshold value;
Step 103, expertise guidance is lower extracts cause and effect topological diagram.
In said program, the analysis quality cause and effect topological diagram correlation of variables further comprises:
Using correlation statistics analysis with machine learning method to the corresponding time series of correlated variables in production process into
Row feature selecting, the d- for generating quality cause and effect topology graph model detach equivalence class;
For d- detach equivalence class in cause and effect topology graph model, using independence test method discrimination variable between because
Fruit direction;
Decomposed using covariance of the related algorithm between the corresponding time series of correlated variables, in conjunction with Granger because
Fruit relationship and theory of statistical test determine the correlation metric between variable;
In conjunction with system operation mechanism and priori, cause and effect direction and correlation metric between the variable are repaiied
Just.
In said program, it is described establish joint data-driven multi-modal monitoring model, further for:
According to the quality cause and effect topology graph model, the prison of the joint data-driven of monitoring quality fault evolution process is established
Model is surveyed, and further establishes the multi-modal monitoring model of joint data-driven, and sets the adaptive of more monitoring models
Process.
It is described to establish the multi-modal monitoring model of joint data-driven in said program, and set more monitoring models
Adaptive process, specifically comprise the following steps:
Step 201, analyze production process in data it is multi-modal;
Step 202, using bayesian theory analyze new data modal idenlification and addition, establish new index of similarity and
Sensitivity index;
Step 203, multi-modal quality cause and effect topological diagram is established to the corresponding data under each mode;
Step 204, consider the quality cause-and-effect diagram correlation metric under single mode;
Step 205, multiple batches of multi-modal quality-monitoring model is established according to the multi-modal quality cause and effect topological diagram;
Step 206, the adaptive process of multiple batches of, multi-modal production process quality-monitoring model is set, and using in fact
Room remote monitoring platform is tested to be verified and tested.
In said program, the Performance Evaluation index that quality fault diagnosis is established based on contribution rate and procedural knowledge, tool
Body includes the following steps:
Step 301, the quality cause and effect topology graph model, multi-modal monitoring model are applied to production process, collect matter
Fault data is measured, fault detect rate, false drop rate and the time-varying characteristics data of quality fault detection are calculated;
Step 302, the lag characteristic detected according to quality fault establishes expected fault detection delay index
(Expected Detection Delay Index, EDDI), introduce formula
Wherein, EDDI is the mathematic expectaion of formula (1), and FDR (Fault Detection Rate) is fault detect rate;
Step 303, the fault detect rate, false drop rate, time-varying characteristics data and time-delay characteristics index are weighted, designs matter
Measure the Performance Evaluation index of fault diagnosis.
In said program, the alignment quality failure specifically comprises the following steps:
Step 401, unified combined monitoring projection subspace and quality fault testing process are established;
Step 402, the propagation path of quality fault, alignment quality failure are identified.
The present invention provides a kind of quality fault localization method based on joint data-driven production process, and this method includes:
Extract quality cause and effect topology graph model;Establish the multi-modal monitoring model of joint data-driven;Based on contribution rate and procedural knowledge
Establish the Performance Evaluation index of quality fault diagnosis;According to the multi-modal monitoring model, quality fault propagation path, root are identified
The Performance Evaluation index alignment quality failure diagnosed according to the quality fault.The present invention is in topological diagram feature extraction, multivariate statistics
On the basis of the process monitoring of data-driven and machine learning, it is proposed that be suitable for the failure of the joint data-driven of quality monitoring
Diagnosis compensates for traditional statistic processes to provide new approach based on the diagnosis of the production process quality fault of data and knowledge
The identification of insoluble quality fault propagation path and fault-location problem are monitored, based on data and knowledge " quantitative-qualitative-
Joint data-driven quantitatively " realizes accurately and efficiently quality fault positioning and diagnosis.
Description of the drawings
Fig. 1 is the hot strip rolling production process technology layout drawing of the embodiment of the present invention 1;
Fig. 2 is the Fault Locating Method Implementation Roadmap based on joint data-driven production process of the embodiment of the present invention 1;
Fig. 3 is the multi-modal quality cause and effect topological diagram of the hot strip rolling rack of the embodiment of the present invention 1.
Specific implementation mode
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with attached drawing and tool
Body embodiment is described in detail.
The present invention is directed to the production process of joint data-driven, it is proposed that a kind of localization method of quality fault is made every effort to gram
Existing data aggregate driving method is taken to the excessively coarse deficiency of process description, the present invention is according to " quantitative-qualitative-quantitative "
Study route, using the associate feature combined data-driven method, have extensively studied between variable based on data with knowledge, accurately
Ground discloses the quality dependent failure i.e. propagation path of quality fault and the source of trouble, realizes the accurate fixed of quality dependent failure
Position, to realize detection early, the diagnosis and maintenance of quality.
With reference to embodiment 1 and attached drawing, the present invention will be further described.
The present embodiment is by taking the hot strip rolling production process that data aggregate drives as an example.It should be noted that the present invention
Fault Locating Method is not limited to hot strip rolling process, is also applied for the production process of other data aggregate drivings, e.g.,
Auto parts machinery production process.
Fig. 1 is the hot strip rolling production process technology layout drawing of the present embodiment.As shown in Figure 1, the present embodiment data join
Close driving hot strip rolling production process, be a typical batch production process, production line include heating furnace, slightly
Milling train group, conveyer belt and flying shear, section cooling, batch unit at mm finishing mill unit.This complicated production process has higher-dimension, non-
Linearly, the characteristics such as time-varying, Coupled Variable, timing dependence, multi-modal, extensive, interval, quality fault is it is possible that in office
What link, and to the monitoring early and judgement of quality fault, then it can ensure being smoothed out for production process.
Fig. 2 is the Fault Locating Method Implementation Roadmap based on joint data-driven production process of the embodiment of the present invention 1.
As shown in Fig. 2, on the basis of the driving of hot strip rolling shown in Fig. 1 this practical engineering application, the present embodiment based on number
Quality fault localization method according to joint driving includes the following steps:
Step S1 extracts quality cause and effect topology graph model.
In the present embodiment, the extraction quality cause and effect topology graph model further extracts to be based on data-driven method
The quality cause and effect topology graph model of hot strip rolling production process designs the quality that suitable topological diagram threshold value realizes data-driven
The trimming and optimization of cause and effect topological diagram, to obtain the quality cause and effect topological diagram of moderate scale.
The extraction process of above topology figure is particularly suitable for complicated batch production process.Specifically, above topology figure carries
Process is taken, is included the following steps:
Step 101, quality of research cause and effect topological diagram correlation of variables analysis method.
First, it is analyzed using correlation statistics corresponding to correlated variables in complicated batch production process with machine learning method
Time series carry out feature selecting, generate quality cause and effect topology graph model d- detach equivalence class;
Secondly, the cause and effect topology graph model in equivalence class is detached for d-, utilizes the independence test such as calculating likelihood score
Cause and effect direction between method discrimination variable;
Then, it is decomposed using covariance of the related algorithm between the corresponding time series of correlated variables, in conjunction with
Granger causalities and theory of statistical test determine the correlation metric between variable;
Finally, in conjunction with prioris such as system operation mechanism, procedural knowledge and expertises, between variable derived above
Cause and effect direction and correlation metric are modified.To which the information propagating pathway for the complicated batch process multivariable of analysis is established
Basis.
Further, it can be realized and be extracted by following procedure:
According to correlation theories such as Granger causalities, the present invention obtains strip heat using Time series analysis method
Causality between tandem rolling variable.
If X and Y is two stochastic variables in hot strip rolling production process, corresponding time series is expressed as
{x1,x2,…,xt}、{y1,y2,…,yt, wherein { xt-k,xt-k+1,…,xt-1}、{yt-k,yt-k+1,…,yt-1Indicated respectively
K periods corresponding history observation is gone, considers following two regression equations:
In formula,The regression fit value in two regression equations is indicated respectively, and L, K indicate y respectivelyt、xtIt is stagnant
Issue afterwards, al、bkFor regression coefficient, εt、ηtFor regression error.If the prediction of the confidence level formula (2) obtained under F statistics
Error is smaller than the prediction error of formula (1), then showing xtWith ytBetween causality be xt→yt。
The present invention is by above xtWith ytBetween Causality Analysis as causal between hot strip rolling manufacturing variables
A kind of theory support, prediction, abnormality detection in time series data etc. play a significant role.But this cause and effect
Relationship depends on historical data unduly during implementation, results in the redundancy even causality of mistake.Therefore, knot of the present invention
The prioris such as collaboration system operation mechanism, process and expertise are related by variable under imperfect information or condition of uncertainty
Property index takes into account, and obtains the causality between accurate variable.
Step 102, between design variable correlation metric threshold value.
Automatically select that there are robusts in unsupervised pattern lower threshold value for during quality cause and effect topological diagram model extraction
The problems such as property is not strong, is analyzed and is located to the alternate data of the corresponding time series of correlated variables in complicated batch production process
Then reason utilizes k nearest neighbor algorithms to estimate the mutual information of these data, using the method for significance test, cohesive process and expert
Knowledge considers that the complicated uncertain factors such as batch process coloured noise and interference, realization have correlation metric under surveillance requirements
Threshold design.
Specifically, the above process can be realized by following procedure:
For to automatically select robust in unsupervised pattern lower threshold value during cause and effect topological diagram model extraction of improving quality
The problems such as property is not strong, the mutual information of these data is had estimated using k nearest neighbor algorithms, using the method for significance test, in conjunction with mistake
Journey and expertise, it is contemplated that the uncertain factors such as complicated batch process coloured noise and interference, realization have under surveillance requirements
The threshold design problem of correlation metric.
Step 103, the cause and effect topological diagram extraction under expertise guidance is realized.
Quality cause and effect topological diagram based on data extraction can have a more redundancy link, and the quality of knowledge based extraction
Cause and effect topological diagram can lead to the missing of a large amount of not intuitive or inessential information.Based on this, the present embodiment devises suitable amendment
Operator, using the prioris such as correlated process knowledge and expertise to the quality cause and effect topological diagram of said extracted carry out trimming with
Optimization, to ensure that the cause and effect topological diagram after trimming and optimizing as the directed acyclic graph of moderate scale, realizes the number under knowledge guidance
According to the Construct question of the quality cause and effect topology graph model of driving.
Step S2 establishes the multi-modal monitoring model of joint data-driven.
In the present embodiment, using multivariate statistics data-driven method to the correlation metric in quality cause and effect topological diagram into
Row time series analysis establishes the monitoring model of the joint data-driven of monitoring quality fault evolution process, while and quality
Relevant model analysis is combined, and establishes the multi-modal monitoring model of a unified joint data-driven, and proposes prison
Survey the adaptive approach of model.
Further, the above process includes the following steps:
Step 201, the multimode analysis of complicated batch process data.
The multimode step response of hot strip rolling production process data is analyzed, and is had using the cluster based on sample geometry
(Between-Within Proportion, BWP) index is divided in the m- class of effect property index-class, is determined in conjunction with clustering algorithm more
The best mode number of mode.
Due to the difference of the production schedule, the change of product index, the change of product category between hot strip rolling process runs
Dynamic, environment variation etc. causes process data to be presented the characteristics such as multiple batches of, multi-modal, dynamic, non-gaussian, multimode analysis according to
According to being that there is similar correlative relationship inside same mode, there is visibly different correlative relationship between different modalities.
In modal characteristics extraction, one side qualitative data (data label) may be incomplete, on the other hand considers that unknown failure may
It is not comprised in training data.In view of the above problems, the present invention uses semi-supervised mixing discriminant analysis and bayesian theory,
The extraction of modal characteristics is completed under procedural knowledge supervision.In to multi-modal processing procedure, the reasonable cluster of training data is (i.e.
Determine best mode number) it is most important to fault detection and diagnosis, therefore the present invention considers from distance measure, introduces a kind of base
In the Cluster Validity Index of sample geometry --- divided in the m- class of class (Between-Within Proportion,
BWP) index:
In formula,Indicate p-th of sample of m classes,Indicate q-th of sample of jth class,Indicate the of jth class
I sample;B (j, i) is defined as the minimum between class distance of i-th of sample of jth class, and w (j, i) defines i-th of sample of jth class
Inter- object distance.By the index, multi-modal best mode number is determined in conjunction with clustering algorithm.
Step 202, new similarity is established using bayesian theory for the modal idenlification of new data and addition problem
Index and sensitivity index calculate new data and belong to each mould probability of state and define a suitable threshold value, if surpassing in long-time
Go out threshold value, so that it may tentatively be judged as new mode, and consider to increase new modal characteristics in mode library.
Bayesian theory is used for the modal idenlification and addition problem, the present invention of the new data of hot strip rolling production,
New index of similarity and sensitivity index are established, new data is calculated and belongs to each mould probability of state and define one suitably
Threshold value, if exceeding the threshold value in for a long time, so that it may tentatively be judged as new mode.
Step 203, on the basis of above-mentioned mode characterization, division are with identification, the corresponding data under each mode is built
Vertical multi-modal quality cause and effect topological diagram.
On the basis of above-mentioned mode characterization, division are with identification, the corresponding data under each mode is established multi-modal
Quality cause and effect topological diagram.
Step 204, consider under single mode quality cause-and-effect diagram correlation metric CI (Correlation Index,
CI it) describes, and only considers associated CI in cause and effect topological diagram, consider two time serieses:P (t)=[CIT(t-1),CIT(t-
2),…]TWith f (t)=[CIT(t),CIT(t+1),…]T, wherein the CI (t) of t moment is m dimensional vectors, utilizes canonical variable point
Analysis (Canonical Variate Analysis, CVA) method analyzes above-mentioned two time series, establishes and is based on quality
The Dynamic monitoring pattern of the correlation metric of propagation.
Step 205, it in conjunction with above-mentioned quality cause and effect topological diagram, is established on the basis of multiple batches of, multi-modal cause and effect topological diagram more
The multi-modal quality-monitoring model of batch.
Step 206, the adaptive technique of multiple batches of, multi-modal hot strip rolling procedure quality monitoring model is studied, and
It is verified and is tested using laboratory remote monitoring platform.
Fig. 3 is the multi-modal quality cause and effect topological diagram of the hot strip rolling rack of the embodiment of the present invention 1.As shown in figure 3, will
Data are divided into several mode by three-dimensional multiple batches of process data by model analysis, established under each mode quality because
Fruit topological diagram.
In front on the basis of research contents, the quality cause-and-effect diagram correlation metric CI under single mode is considered
(Correlation Index, CI) is described, and only considers associated CI in cause and effect topological diagram, it is contemplated that the following two time
Sequence:P (t)=[CIT(t-1),CIT(t-2),…]TWith f (t)=[CIT(t),CIT(t+1),…]T, wherein the CI of t moment
(t) it is m dimensional vectors, when using canonical variate analysis (Canonical Variate Analysis, CVA) method to above-mentioned two
Between sequence analyzed, establish the Dynamic monitoring pattern of the correlation metric based on mass propagation.In conjunction with above-mentioned quality because
Fruit topological diagram then can establish multiple batches of, multi-modal quality-monitoring model on the basis of multiple batches of, multi-modal cause and effect topological diagram.
Step S3 establishes the Performance Evaluation index of quality fault diagnosis based on contribution rate and procedural knowledge.
In the present embodiment, the relevant performance of fault diagnosis evaluation index of quality, and the joint number to being proposed are defined
According to driving hot strip rolling quality fault propagation path identification and Fault Locating Method and conventional fault diagnosis method from matter
Hysteresis quality, fault detect rate and the false drop rate etc. of amount dependent failure detection are assessed, and are diagnosed to the quality dependent failure of proposition
Method has carried out quantitative analysis and evaluation, after improved and raising, it is determined that laboratory test milling train and commercial Application authentication
Case completes the application of hot strip rolling production scene.
Step S3 further comprises following steps:
Step 301, by the main theory achievement obtained in the above research contents and algorithm hot strip rolling production line into
Row application verification will not only consider traditional fault detect rate and mistake for the evaluation index of quality dependent failure diagnosis performance
Inspection rate, and to consider time-varying characteristics failure.
Step 302, for the lag characteristic of quality dependent failure detection, it is proposed that a new expection fault detection delay
Property index (Expected Detection Delay Index, EDDI), introduce formula
Wherein, EDDI is the mathematic expectaion of formula (4), and FDR (Fault Detection Rate) is fault detect rate.
Step 303, the above evaluation index is weighted, designs the Performance Evaluation of the relevant method for diagnosing faults of final quality
Index.
Step 304, laboratory test milling train and commercial Application proof scheme are determined, hot strip rolling production scene is completed
Using achieving good effect.
Step S4 identifies quality fault propagation path, is examined according to the quality fault according to the multi-modal monitoring model
Disconnected Performance Evaluation index alignment quality failure.
In the present embodiment, the detection of quality dependent failure, fault propagation path identification and event of joint data-driven are realized
Hinder localization method, steps are as follows for specific implementation:
Step 401, unified combined monitoring projection subspace and the relevant fault detection method of quality are established.
After obtaining multiple batches of multi-modal quality-monitoring model, so that it may online to be carried out to multi-modal complicated batch process
Monitoring and fault diagnosis.Since complicated batch process is each stablized, mode run time is shorter, sampled data is also less, stage die
State run time is shorter, sampled data is less, and often to be switched between different modalities.Such conventional monitoring methods
The assumed condition of Gaussian Profile on batch direction is cannot be satisfied, it is slowly varying that potentially relevant characteristic on batch direction can not be met
Engine request, cannot be satisfied the hypotheses of normal distribution on batch direction, also cannot be satisfied potentially relevant on batch direction
The actual requirement that characteristic slowly fluctuates, just seeming cannot be applicable in, ineffective.In view of the above problems, the present invention proposes one kind
Joint fault detection method based on CVA establishes unified prison using the integrated information of above-mentioned multi-modal quality-monitoring model
Model is surveyed, the real-time monitoring of multi-modal data is realized, avoids the frequent switching of monitoring model.
Step 402, the propagation path identification of quality dependent failure and Fault Locating Method.
Using process and expertise as prior probability, knowledge is dissolved into relative contribution rate using bayesian theory,
The propagation path that quality dependent failure is recognized with certain search strategy is classified into two kinds of situation difference of single failure and multiple faults
Carry out correlative study, the final propagation path identification for solving multiple faults, to realize fault location.
Indicate that each correlation metric (failure path) to the influence degree of Testing index, fully diagnoses matter for clarity
Relevant failure is measured, the present invention realizes the relevant fault diagnosis of quality using relative contribution rate method.The specific steps are:Conventional method
Typically utilize T2The relevant failure of statistical monitoring quality, Q statistical monitoring process noises, but since the variation of process noise may
Product quality variable can be influenced, therefore can be using Q as T2Supplement, that is, use T2And the Testing index φ detections quality of Q synthesis is relevant
Failure, the probability density function by calculating φ acquire control limit, and terraced from the first order Taylor of φ functions and kernel function
Degree sets out, and studies each correlation metric to the influence degree of Testing index φ, as the relative contribution rate of quality dependent failure.
Meanwhile knowledge is dissolved into phase by the present invention using process and expertise as prior probability using bayesian theory
To in contribution rate, the propagation path of quality dependent failure being recognized with certain search strategy, is classified into single failure and multiple faults
Two kinds of situations have carried out correlative study respectively, finally solve the problems, such as the propagation path identification of multiple faults, to realize failure
Positioning.
The present invention propose it is a kind of joint data-driven hot strip rolling quality fault propagation path identification and failure
Localization method, this method include:Based on production process data and process knowledge, it is proposed that the quality cause and effect of complicated batch process is opened up
Flutter figure extracting method;Establish the relevant malfunction monitoring model of complicated batch process quality of joint data-driven;Construct base
In the relevant performance of fault diagnosis analysis and evaluation index of complicated batch process quality of joint data-driven;Establish joint number
According to the relevant malfunction monitoring of hot strip rolling procedure quality of driving, the integrated frame of fault propagation Path Recognition and fault location
Frame provides a set of new technology and solution for the complexity relevant fault diagnosis of batch process quality.
The present embodiment will be in the process monitoring of the topological diagram feature extraction of existing data-driven, multivariate statistics data-driven
In technology and the theoretical research foundation of machine learning, it is proposed that be suitable for the fault diagnosis skill of the joint data-driven of quality monitoring
Art.Fault Locating Method provided by the present invention based on joint data-driven production process, will be for based on data and knowledge
The relevant fault diagnosis of hot strip rolling production process quality provides new thinking and approach, compensates for traditional statistic processes prison
Insoluble quality dependent failure propagation path identification and fault location and the diagnosis problem of multiple faults are controlled, the present invention proposes
The data-driven method of combining of " quantitative-qualitative-quantitative " based on data and knowledge provided for the relevant fault diagnosis of quality
New technology and means.
The above is the preferred embodiment of the present invention, it is noted that for those skilled in the art
For, without departing from the principles of the present invention, several improvements and modifications can also be made, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of quality fault localization method driving production process based on data aggregate, which is characterized in that the method includes
Following steps:
Extract quality cause and effect topology graph model;
Establish the multi-modal monitoring model of joint data-driven;
The Performance Evaluation index of quality fault diagnosis is established based on contribution rate and procedural knowledge;
According to the multi-modal monitoring model, quality fault propagation path is identified, the performance diagnosed according to the quality fault is commented
Estimate index alignment quality failure.
2. Fault Locating Method according to claim 1, which is characterized in that the extraction quality cause and effect topology graph model,
It is further that the procedural knowledge driven based on data aggregate extracts quality cause and effect topology graph model with historical data.
3. Fault Locating Method according to claim 2, which is characterized in that the extraction quality cause and effect topology graph model,
Specifically comprise the following steps:
Step 101, quality cause and effect topological diagram correlation of variables is analyzed;
Step 102, between design variable correlation metric threshold value;
Step 103, expertise guidance is lower extracts cause and effect topological diagram.
4. Fault Locating Method according to claim 3, which is characterized in that the analysis quality cause and effect topological diagram variable phase
Closing property further comprises:
Spy is carried out to the corresponding time series of correlated variables in production process with machine learning method using correlation statistics analysis
Sign selection, the d- for generating quality cause and effect topology graph model detach equivalence class;
Cause and effect topology graph model in equivalence class is detached for d-, utilizes the cause and effect side between the method discrimination variable of independence test
To;
It is decomposed using covariance of the related algorithm between the corresponding time series of correlated variables, is closed in conjunction with Granger causes and effects
System and theory of statistical test, determine the correlation metric between variable;
In conjunction with system operation mechanism and priori, cause and effect direction and correlation metric between the variable are modified.
5. Fault Locating Method according to claim 1, which is characterized in that described to establish the multi-modal of joint data-driven
Monitoring model, further for:
According to the quality cause and effect topology graph model, the monitoring mould of the joint data-driven of monitoring quality fault evolution process is established
Type, and the multi-modal monitoring model of joint data-driven is further established, and set the adaptive process of more monitoring models.
6. Fault Locating Method according to claim 5, which is characterized in that described to establish the multi-modal of joint data-driven
Monitoring model, and the adaptive process of more monitoring models is set, specifically comprise the following steps:
Step 201, analyze production process in data it is multi-modal;
Step 202, modal idenlification and the addition that new data is analyzed using bayesian theory, establish new index of similarity and sensitive
Spend index;
Step 203, multi-modal quality cause and effect topological diagram is established to the corresponding data under each mode;
Step 204, consider the quality cause-and-effect diagram correlation metric under single mode;
Step 205, multiple batches of multi-modal quality-monitoring model is established according to the multi-modal quality cause and effect topological diagram;
Step 206, the adaptive process of multiple batches of, multi-modal production process quality-monitoring model is set, and utilizes laboratory
Remote monitoring platform is verified and is tested.
7. Fault Locating Method according to claim 1, which is characterized in that described to be established based on contribution rate and procedural knowledge
The Performance Evaluation index of quality fault diagnosis, specifically comprises the following steps:
Step 301, the quality cause and effect topology graph model, multi-modal monitoring model are applied to production process, collect quality event
Hinder data, calculates fault detect rate, false drop rate and the time-varying characteristics data of quality fault detection;
Step 302, the lag characteristic detected according to quality fault establishes expected fault detection delay index (Expected
Detection Delay Index, EDDI), introduce formula
Wherein, EDDI is the mathematic expectaion of formula (1), and FDR (Fault Detection Rate) is fault detect rate;
Step 303, the fault detect rate, false drop rate, time-varying characteristics data and time-delay characteristics index are weighted, designing quality event
Hinder the Performance Evaluation index of diagnosis.
8. Fault Locating Method according to claim 1, which is characterized in that the alignment quality failure, specifically include as
Lower step:
Step 401, unified combined monitoring projection subspace and quality fault testing process are established;
Step 402, the propagation path of quality fault, alignment quality failure are identified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610391112.3A CN105955241B (en) | 2016-06-03 | 2016-06-03 | A kind of quality fault localization method based on joint data-driven production process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610391112.3A CN105955241B (en) | 2016-06-03 | 2016-06-03 | A kind of quality fault localization method based on joint data-driven production process |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105955241A CN105955241A (en) | 2016-09-21 |
CN105955241B true CN105955241B (en) | 2018-09-14 |
Family
ID=56907649
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610391112.3A Active CN105955241B (en) | 2016-06-03 | 2016-06-03 | A kind of quality fault localization method based on joint data-driven production process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105955241B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107402559B (en) * | 2017-08-21 | 2019-05-24 | 北京化工大学 | A kind of batch process measurement data method for detecting abnormality based on the variation of dynamic Hyper-sphere Structure |
CN108664002B (en) * | 2018-04-27 | 2021-06-29 | 中国石油大学(华东) | Quality-oriented nonlinear dynamic process monitoring method |
CN108762238B (en) * | 2018-06-15 | 2020-07-28 | 东北大学 | DCD-based hydrometallurgy leaching process fault diagnosis method |
CN109471420B (en) * | 2018-09-21 | 2020-08-14 | 浙江大学 | CVA-SFA-based method for monitoring control performance of air preheater of large coal-fired power generator set of intelligent power plant |
DE102019209110A1 (en) * | 2019-06-24 | 2020-12-24 | Sms Group Gmbh | Industrial plant, in particular plant in the metal-producing industry or the aluminum or steel industry, and method for operating an industrial plant, in particular a plant in the metal-producing industry or the aluminum or steel industry |
CN111695583A (en) * | 2019-07-18 | 2020-09-22 | 广东电网有限责任公司信息中心 | Feature selection method based on causal network |
CN111367253B (en) * | 2020-02-18 | 2021-03-16 | 清华大学 | Chemical system multi-working-condition fault detection method based on local adaptive standardization |
CN111290369A (en) * | 2020-02-24 | 2020-06-16 | 苏州大学 | Fault diagnosis method based on semi-supervised recursive feature retention |
CN111914887B (en) * | 2020-06-13 | 2024-02-09 | 广东越凯新材料有限公司 | Novel multi-mode chemical process abnormal state detection method |
CN111537945B (en) * | 2020-06-28 | 2021-05-11 | 南方电网科学研究院有限责任公司 | Intelligent ammeter fault diagnosis method and equipment based on federal learning |
CN111722043B (en) * | 2020-06-29 | 2021-09-14 | 南方电网科学研究院有限责任公司 | Power equipment fault detection method, device and system |
CN113835411B (en) * | 2021-09-07 | 2023-10-27 | 北京科技大学顺德研究生院 | Comprehensive diagnosis method for quality abnormality of steel rolling process flow |
CN114077876B (en) * | 2022-01-19 | 2022-05-13 | 北京科技大学 | Strip steel hot continuous rolling multi-mode process monitoring method and device |
CN115780530B (en) * | 2023-01-19 | 2023-05-09 | 北京科技大学 | Steel rolling fault tracing method and device based on expert knowledge and data combined driving |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3320101B2 (en) * | 1992-07-20 | 2002-09-03 | キヤノン株式会社 | Image forming device |
CN101446827A (en) * | 2008-11-06 | 2009-06-03 | 西安交通大学 | Process fault analysis device of process industry system and method therefor |
CN101634851A (en) * | 2009-08-25 | 2010-01-27 | 西安交通大学 | Method based on cause-and-effect relation of variables for diagnosing failures in process industry |
CN102765643A (en) * | 2012-05-31 | 2012-11-07 | 天津大学 | Elevator fault diagnosis and early-warning method based on data drive |
CN103869739A (en) * | 2012-12-11 | 2014-06-18 | 新昌县冠阳技术开发有限公司 | Algorithm improvement and software system based on monitoring of data driving process |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3805993B2 (en) * | 2001-03-30 | 2006-08-09 | 株式会社東芝 | Cause estimation method, cause estimation device, cause estimation program, and recording medium |
CN101783749B (en) * | 2009-12-24 | 2011-04-27 | 北京市天元网络技术股份有限公司 | Network fault positioning method and device |
US20130197698A1 (en) * | 2012-01-26 | 2013-08-01 | Carrier Corporation | HVAC System Fault Root Cause Self-Determination |
CN104699050A (en) * | 2015-02-13 | 2015-06-10 | 浙江中烟工业有限责任公司 | Leaf-shred preparation segment on-line monitoring and fault diagnosing method for cigarette filament treatment driven by data |
-
2016
- 2016-06-03 CN CN201610391112.3A patent/CN105955241B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3320101B2 (en) * | 1992-07-20 | 2002-09-03 | キヤノン株式会社 | Image forming device |
CN101446827A (en) * | 2008-11-06 | 2009-06-03 | 西安交通大学 | Process fault analysis device of process industry system and method therefor |
CN101634851A (en) * | 2009-08-25 | 2010-01-27 | 西安交通大学 | Method based on cause-and-effect relation of variables for diagnosing failures in process industry |
CN102765643A (en) * | 2012-05-31 | 2012-11-07 | 天津大学 | Elevator fault diagnosis and early-warning method based on data drive |
CN103869739A (en) * | 2012-12-11 | 2014-06-18 | 新昌县冠阳技术开发有限公司 | Algorithm improvement and software system based on monitoring of data driving process |
Non-Patent Citations (5)
Title |
---|
Dynamic Uncertain Causality Graph for Knowledge Representation and Reasoning: Continuous Variable, Uncertain Evidence, and Failure Forecast;Qin Zhang;《IEEE Transactions on Systems, Man, and Cybernetics: Systems》;20150217;第45卷(第7期);第990-1003页 * |
Online Contribution Rate Based Fault Diagnosis for Nonlinear Industrial Processes;PENG Kai-Xiang;《ACTA AUTOMATICA SINICA》;20140331;第40卷(第3期);第423-430页 * |
基于数据驱动的大型化工装置健康评估与故障源诊断;曹稼斌 等;《计算机集成制造系统》;20130531;第19卷(第5期);第1085-1090页 * |
数据驱动的多变量控制系统性能监测与诊断;陈贵;《万方数据库》;20131008;第1-138页 * |
质量相关的带钢热连轧过程监控;彭开香 等;《控制工程》;20110731;第18卷(第4期);第650-654页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105955241A (en) | 2016-09-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105955241B (en) | A kind of quality fault localization method based on joint data-driven production process | |
He et al. | A fuzzy TOPSIS and rough set based approach for mechanism analysis of product infant failure | |
CN101963802B (en) | Virtual measurement method in batch manufacture procedure and system therefor | |
CN103345593B (en) | The gathering method for detecting abnormality of macro cell facing single data stream | |
CN103777627B (en) | A kind of batch process on-line monitoring method based on a small amount of batch | |
CN110110870A (en) | A kind of equipment fault intelligent control method based on event graphical spectrum technology | |
CN111737909B (en) | Structural health monitoring data anomaly identification method based on space-time graph convolutional network | |
CN109766583A (en) | Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique | |
CN109685366A (en) | Equipment health state evaluation method based on mutation data | |
Ma et al. | A novel hierarchical detection and isolation framework for quality-related multiple faults in large-scale processes | |
CN103116306B (en) | Automatic stepping type ordered time interval dividing method | |
Li et al. | A novel diagnostic and prognostic framework for incipient fault detection and remaining service life prediction with application to industrial rotating machines | |
CN102789545A (en) | Method for predicating remaining life of turbine engine based on degradation model matching | |
CN105629958B (en) | A kind of batch process method for diagnosing faults based on sub-period MPCA SVM | |
CN104899327A (en) | Method for detecting abnormal time sequence without class label | |
CN105511445A (en) | Multi-modal process fault detection method based on local neighbor standardization matrix | |
CN104471542A (en) | Discriminative hidden kalman filters for classification of streaming sensor data in condition monitoring | |
CN110757510B (en) | Method and system for predicting remaining life of robot | |
Dong et al. | A novel fault detection method based on the extraction of slow features for dynamic nonstationary processes | |
CN110083860A (en) | A kind of industrial method for diagnosing faults based on correlated variables selection | |
CN105334185A (en) | Spectrum projection discrimination-based near infrared model maintenance method | |
CN104182623A (en) | Thermal process data detection method based on equivalent change rate calculation | |
CN103389701A (en) | Plant-level process fault detection and diagnosis method based on distributed data model | |
Ferrer et al. | PLS: A versatile tool for industrial process improvement and optimization | |
CN109116319B (en) | Fault detection method for radar system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |