CN102262690B - Modeling method of early warning model of mixed failures and early warning model of mixed failures - Google Patents
Modeling method of early warning model of mixed failures and early warning model of mixed failures Download PDFInfo
- Publication number
- CN102262690B CN102262690B CN 201110150703 CN201110150703A CN102262690B CN 102262690 B CN102262690 B CN 102262690B CN 201110150703 CN201110150703 CN 201110150703 CN 201110150703 A CN201110150703 A CN 201110150703A CN 102262690 B CN102262690 B CN 102262690B
- Authority
- CN
- China
- Prior art keywords
- analysis module
- variable
- reasoning
- warning model
- module
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The embodiment of the invention provides a modeling method of an early warning model of mixed failures and a modeling system. The modeling method provided by the invention comprises the following steps of: generating a function analyzing module on the basis of HAZOP (Hazard and Operability Analysis) or FMEA (Failure Mode and Effects Analysis); generating a degeneration analyzing module on the basis of FMEA analyzing results and a theory of stochastic processes; generating an accident analyzing module according to state monitoring data and maintenance action information; generating an action analyzing module according to output results of the function analyzing module and the degeneration analyzing module through combining a DBN (Dynamic Bayesian Network) theory; taking the output of the accident analyzing module as an inference evidence and utilizing a DBN inference algorithm to process forward and backward inferences in the same time period to generate an evaluating module for outputting factors and consequences of system failures; taking the output results of the evaluating module and the accident analyzing module as the inference evidence and utilizing the DBN inference algorithm to process forward and backward inferences in the different time periods to generate a predicating module for outputting prospective degeneration tendencies of each member of the system. The model provided by the invention can be used for tracking the failure factors of the system and inferring possible failure consequences and probability.
Description
Technical field
The present invention relates to the safety engineering field, relate to particularly a kind of modeling method and modeling of mixed fault Early-warning Model.
Background technology
Along with the development of Condition Monitoring Technology, increasing experts and scholars are devoted to the detection and classification of Study system fault, and have developed the condition monitoring and diagnosis software and hardware system of many maturations.But " all kinds of diseases and ailments are anti-not as one " in order to improve the essential safety of system, must fundamentally avoid by fault pre-alarming the condition of fault generation.Existing method is often paid close attention to the degradation mechanism of single parts or an independent particle system and the research of residual life.Yet, complication system scarcely is to move in the such stable environment in laboratory, but be subject to the impact of various inside and outside random occurrences, mutual, environmental evolution between the system, human error, machine ages and other uncertain factors all can cause generation and the evolution of faults coupling effect, and so that the behavior of system has randomness.Although simply it is split as the difficulty that unit has independently reduced problem, can not get optimum solution, do not meet the Practical Project situation yet.Especially when system exists common cause failure (common cause failure), relevant failure or parts to have the situation of Multiple Failure Modes, existing forecast model has been ignored between the system variable and the interaction between the malfunction and failure pattern and influence degree, the rationality of reasoning and decision-making is lower, and produces easily wrong report or fail to report phenomenon.
All parts fault mode self and interactional development and change process are presented as a fault chain of causation in the complication system, and it will be associated by a series of fault modes that initial fault mode causes by cause-effect relationship.The faults coupling effect of complication system is actually one take the network of the fault chain of causation as the basis, the fault that one or a few node occur (may be to occur at random, also may be that human factor causes) can cause that other nodes break down, and finally cause the collapse of quite a few node even whole network by the coupled relation between the network node.Wherein, the fault chain of causation refers to: all parts fault mode self and interactional development and change process are presented as a fault chain of causation in the complication system, and it will be associated by a series of fault modes that initial fault mode causes by cause-effect relationship.
Therefore, the condition for fear of fault produces improves the rationality of fault rootstock sexual factor identification, demands working out a kind of mixed fault Early-warning Model based on the fault chain of causation that is achieved as follows function urgently:
(1) wishing can be before breaking down, and carries out the identification of root danger by the mixed fault Early-warning Model, takes Pre-control measures to make system remain on specified states;
(2) when early stage Single Point of Faliure occurs, hope can be predicted the development trend that broken down by the mixed fault Early-warning Model and on the impact of other single-point states, be conducive to adopt initiatively disengagement failure travel path of fault isolation, prevent fault pervasion and cause the destruction of other single-point states even the collapse of system.
Summary of the invention
The object of the invention is to, remedy the deficiency of domestic existing complication system fault pre-alarming model, a kind of accurate, reasonable, effective mixed fault Early-warning Model is provided, quantitative modeling and reasoning by the fault chain of causation, break through the required assumed conditions such as component failure independence of classic method, effectively realize fault pre-alarming analyze in to the accurate identification of multi-part, many dangerous matter sources system failure root sexual factor, and to the reasonable prediction of the following degradation trend of parts and remaining life thereof.
On the one hand, the embodiment of the invention provides a kind of modeling method of mixed fault Early-warning Model, and described method comprises: analyze HAZOP or failure mode and effect analysis FMEA, systematic function analysis module based on hazard and operability; Based on FMEA analysis result and discrete time Markov theory of random processes, generate the degradation analysis module; According to Condition Monitoring Data and maintenance information, generate event analysis module; Output rusults according to described functional analysis module and described degradation analysis module is theoretical in conjunction with dynamic bayesian network, generates the behavioural analysis module; Take the Observable state variable Real-Time Monitoring value of event analysis module output as the reasoning evidence, utilize the dynamic bayesian network reasoning algorithm in the same timeslice of described behavioural analysis module, to carry out front and back to reasoning, generate evaluation module, the possible consequence that output system failure factor and fault cause; Take the Observable state variable Real-Time Monitoring value of the Output rusults of described evaluation module and the output of described event analysis module as the reasoning evidence, utilize between the different timeslice of dynamic bayesian network reasoning algorithm and carry out front and back to reasoning, the generation forecast module, the degradation trend in output system all parts future.
On the other hand, the embodiment of the invention provides a kind of modeling of mixed fault Early-warning Model, comprise: the functional analysis module, be used for determining dynamic node and the static node of mixed fault Early-warning Model, the implicit state variable that dynamic node is corresponding and the state space of implicit state variable, the observational variable that static node is corresponding and the state space of observational variable, the failure cause related with each node and consequence, and the incidence relation between failure cause and the consequence; The degradation analysis module is used for component failure data and priori according to historical data base, determines state transitions rule and the failure probability density function of the implicit state variable that each dynamic node of mixed fault Early-warning Model is corresponding; Event analysis module is used for real time data and the chronic frustration data that monitor are stored into described historical data base, for the reasoning process of evaluation module and prediction module provides the reasoning evidence; The behavioural analysis module is used for setting up network structure and the parameter of mixed fault Early-warning Model, utilizes simultaneously Condition Monitoring Data, fail data in the described historical data base parameter of described mixed fault Early-warning Model is estimated and to be upgraded; Evaluation module is used for implicit state, system failure root sexual factor, dangerous reason at different levels, dangerous consequences and the corresponding safety practice through reasoning output current system all parts, at least one of turnaround plan; Prediction module is used at least one according to each observed parameter of the implicit State-output system of described all parts variate-value in future, the parts degradation trend in future, remaining life, predictive maintenance strategy.
The technique scheme that the embodiment of the invention provides, a series of by making up " functional analysis modules ", " degradation analysis module ", " behavioural analysis module ", " event analysis module " also merge it with structure mixed fault Early-warning Model, and the device systems fault chain of causation is carried out the quantification modeling.The mixed fault Early-warning Model that builds can be according to the observational variable value of " event analysis module " real-time storage, at one time in the sheet between (" Spatial Dimension ") and the different timeslice (" time dimension ") carry out front and back to reasoning, realization to the diagnosis of system failure root sexual factor, fault may consequence the prediction of prediction, system degradation trend and preventative maintenance the establishment of the project etc., the essential safety of safeguards system.
Description of drawings
In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, the below will do one to the accompanying drawing of required use in embodiment or the description of the Prior Art and introduce simply, apparently, accompanying drawing in the following describes only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain according to these accompanying drawings other accompanying drawing.
Fig. 1 is the mixed fault Early-warning Model construction and application Organization Chart of the embodiment of the invention;
Fig. 2 is the modeling process flow diagram of the mixing Early-warning Model of the embodiment of the invention;
Fig. 3 a is the east of a lake, the Qinghai compressor station Compressor Set entire system synoptic diagram of the embodiment of the invention;
Fig. 3 b is the axial-flow compressor of the east of a lake, the Qinghai compressor station Compressor Set system of the embodiment of the invention;
Fig. 3 c is the gas-turbine combustion chamber of the east of a lake, the Qinghai compressor station Compressor Set system of the embodiment of the invention;
Fig. 3 d is the gas turbine backup system of the east of a lake, the Qinghai compressor station Compressor Set system of the embodiment of the invention;
Fig. 4 is the surface chart of Compressor Set " event analysis module " application example of the embodiment of the invention;
Fig. 5 is the mixed fault Early-warning Model schematic network structure of the oil system of the embodiment of the invention;
Fig. 6 is the generation of the embodiment of the invention reliability development trend synoptic diagram in oil system future of degenerating;
Fig. 7 is the degradation trend synoptic diagram that affects lower bearing in the oil system of degenerating of the embodiment of the invention;
Fig. 8 is the following reliability development trend of oil system synoptic diagram after the maintenance of the embodiment of the invention;
Fig. 9 a is the air system degeneration of the embodiment of the invention and implies state variable future trends synoptic diagram;
Fig. 9 b is that the air system of the embodiment of the invention is degenerated and observed parameter variable future trends synoptic diagram;
Figure 10 a is that the air system under the given preventative maintenance scheme of the embodiment of the invention is degenerated and implied state variable future trends synoptic diagram;
Figure 10 b is that the air system under the given preventative maintenance scheme of the embodiment of the invention is degenerated and observed parameter variable future trends synoptic diagram.
Embodiment
For the purpose, technical scheme and the advantage that make the embodiment of the invention clearer, below in conjunction with the accompanying drawing in the embodiment of the invention, technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that obtains under the creative work prerequisite.
The embodiment of the invention provides a kind of mixed fault Early-warning Model and modeling method thereof take dynamic bayesian network as theoretical foundation, and the fault chain of causation is carried out quantification modeling and reasoning.From many chains of causation because of many fruits, find out the most initial reason that produces failure symptom; And further trouble-shooting travel path, the future developing trend that broken down of prediction and to the negative effect of other single-point states, thereby strengthen the rationality to the identification of fault rootstock sexual factor, the condition of avoiding fault to produce, the basic goal of realization fault pre-alarming.The embodiment of the invention on the basis of having merged system architecture/hardware, process function, historical fail data and Condition Monitoring Data, the inherent evolution process of executive system and parts self and the each other abstract and modeling of reciprocal process.It comprises 6 modeling process, and Fig. 1 is the modeling of mixed fault Early-warning Model and the application architecture synoptic diagram of the embodiment of the invention, and Fig. 2 is the modeling process flow diagram of the mixed fault Early-warning Model of the embodiment of the invention.In conjunction with consulting Fig. 1 and Fig. 2, the specific implementation process comprises the steps:
As shown in Figure 1, one embodiment of the present of invention need P﹠amp in the building process of functional analysis module; ID schemes (technique and instrument process flow diagram, Process﹠amp; Instrument Diagram) and expertise.It is a kind of a kind of method ready-made and that generally use in petrochemical complex safety analysis field that HAZOP analyzes.HAZOP analyzes and adopts standardization " introducer " that system's intermediate variable is set " departing from ", along " departing from " reverse find in system improper " reason ", and along " departing from " forward lookup in system unfavorable " consequence ".For the existing safety practice of each important " reason-consequence " antithesis, rectification or new safety practice suggestion are proposed in the seeking system.Wherein, " reason-consequence " antithesis refers to reason and the dangerous consequence that causes of initiation potential.In the present embodiment, the production run of apparatus system is the process that makes material generation predetermined physical and chemical change by unit operations and physical-chemical reaction.In order to prevent runaway reaction and industrial accident, these variations (representing mainly with technological parameter) must strictly be controlled, can not depart from predetermined scope and form, this class technological parameter be departed from predetermined scope and form is called " departing from " (comprise quantity of state departs from, observed quantity depart from).
FMEA analyzes and to refer to: can be divided into the characteristics of subsystem, equipment and element according to system, according to actual needs system be cut apart, then analyze the impact of contingent fault type separately and generation thereof, in order to take Counter-measures.
Further, analyze about HAZOP and FMEA, those skilled in the art can be with reference to following document (wherein having detailed possible dangerous reason and the analytical procedure of consequence): [1] Shao Hui, " system safety engineering ", petroleum engineering publishing house, 2008.5.
Wherein, above-mentionedly analyze and " the functional analysis module " that generate based on HAZOP and FMEA, specifically can be used for: determine the key parameter of mixed fault Early-warning Model, comprise dynamic node and static node in the mixed fault Early-warning Model; And, determine variable and parameter space thereof that each node is corresponding, corresponding implicit state variable and the state space thereof of dynamic node for example, the observational variable that static node is corresponding and state space thereof; And, reach the reciprocal effect relation between each parts and the environment between all parts.Wherein, state space refers to the set of all possible states.Parameter space comprises the probability that state space and each state occur.
In the present embodiment, this degradation analysis module, each failure mode quantification ground that can be used for FMEA is analyzed is abstract to be the implicit state variable corresponding to dynamic node of mixed fault Early-warning Model, and according to fail data and priori, determine the model parameter such as state transitions rule, priori failure probability of each implicit state variable.
Wherein, above-mentioned discrete time Markov stochastic process theory those skilled in the art can consult following list of references: Zhou Mengqing, " theory of random processes ", Electronic Industry Press, 2009.4.
Particularly, above-mentioned Condition Monitoring Data can comprise from spot sensor, PLC(Programmable logic Controller, Programmable Logic Controller) or SCADA(Supervisory Control And Data Acquisition, data acquisition with monitor control) Real-time Monitoring Data and various Action Events record that system obtains, these information are stored into historical data base.Above-mentioned maintenance information can comprise maintenance record or various preventive maintenance schedule.
In the present embodiment, this event analysis module can be stored into historical data base with the data that monitor, so that for the performed parameter estimation procedure of behavioural analysis module described later provides continuous renewal basic with perfect data, this process is to ensure accurate, the reliable basis of fault pre-alarming analysis.
Particularly, the concrete processing procedure of step 106 can comprise: in time evolution process and the reciprocal process between the variable of implicit state variable, the observed parameter variable of the mixed fault Early-warning Model of " functional analysis module " and " degradation analysis module " output, above-mentioned each variable self inherence is merged (concrete fusion process hereinafter has detailed description) consist of " behavioural analysis module "; Data in the recycling historical data base (for example fail data, historical observation parametric variable value) adopt the DBN parameter estimation algorithm that the parameter of " behavioural analysis module " is estimated.
Wherein, above-mentioned DBN parameter estimation algorithm those skilled in the art can consult following list of references: Xiao Qinkun, Gao Song, high twilight, " the dynamic bayesian network reasoning theories of learning and application ", Beijing: National Defense Industry Press, 2007.
In the present embodiment, " evaluation module " of mixed fault Early-warning Model utilizes the dynamic bayesian network reasoning algorithm to carry out front and back to reasoning (Forward and backward reasoning algorithm) in the same timeslice of behavior analysis module, take " event analysis module " output Observable state variable Real-Time Monitoring value as the reasoning evidence, from " Spatial Dimension " upper each dynamic node of derivation mixed fault Early-warning Model corresponding implicit state variable and corresponding probability of happening, thereby to the implicit state of the current all parts of terminal output system, the fault rootstock sexual factor, dangerous reasons at different levels, possible dangerous consequences, in safety practice and the turnaround plan at least one.
The Observable state variable Real-Time Monitoring value of the Output rusults of step 110, above commentary valency module and the output of above-mentioned event analysis module is the reasoning evidence, utilize the dynamic bayesian network reasoning algorithm between different timeslices, to carry out front and back to reasoning, the generation forecast module is with the output system all parts degradation trend in future.
In the present embodiment, the result of " prediction module " combination " evaluation module " output of mixed fault Early-warning Model (as, implicit state variable) and the Observable state variable Real-Time Monitoring value of " event analysis module " output be the reasoning evidence, utilize dynamic bayesian network to carry out time slice continuation (" time dimension "), carry out front and back to reasoning between the different timeslice of the behavioural analysis module of mixed fault Early-warning Model, prediction obtains component degradation until the temporal evolution process that lost efficacy.Thereby to the terminal output system observed parameter variate-value in future, implicit state variable value, the reliability variation tendency in future etc., accurate for formulating, rational preventive maintenance Best Times or gap periods provide foundation.
Wherein, above-mentioned time slice extensional correlation procedure concept those skilled in the art can consult following list of references: Xiao Qinkun, Gao Song, high twilight, " the dynamic bayesian network reasoning theories of learning and application ", Beijing: National Defense Industry Press, 2007.
The below is described in further detail embodiment of the invention method shown in Figure 2.
(1) modeling process of " functional analysis module " in the mixed fault Early-warning Model:
" functional analysis module " analyzed based on HAZOP and FMEA, determine the key parameter in the mixed fault Early-warning Model, comprise the dynamic node in the mixed fault Early-warning Model, static node, the variable that each node is corresponding and parameter space thereof, and between all parts and and environment between reciprocal effect relation.
The fundamental purpose of " functional analysis module " is that the qualitative cause-effect relationship between the system failure is researched and analysed, and sets up the relational model that influences each other (reciprocal process) between the system unit.Preferably, the embodiment of the invention adopts a kind of two-way function-failure analysis mechanism (dual functioning – malfunctioning reasoning).
(2) modeling process of " degradation analysis module " in the mixed fault Early-warning Model:
" degradation analysis module " determines implicit state variable and the model parameters such as state transitions rule, priori failure probability thereof that dynamic node is corresponding in the mixed fault Early-warning Model.
At first from the output of " functional analysis module ", select crucial failure mode to set up degenerative process, the correlativity between the then identification degenerative process.Crucial failure mode wherein, referring on the ordinary meaning has safely the failure mode that has a strong impact on to parts.Can be come the degenerative process of system core is carried out modeling by discrete time markov stochastic process (DTMP), specific algorithm be as follows:
A degenerative process { X independently
kCan be by its discrete state space χ
XTransition matrix P with correspondence
XRepresent.For the modeling of stochastic dependence (dependent) degenerative process, the embodiment of the invention adopts a kind of short-cut method, will have interactional several process to integrate with in the single model.For example, with two relevant degenerative process { A
t, { B
tMerge into " Macroscopic Process " { AB
t, this merging mode is based on DTMP, and just like giving a definition:
1, merging phase space χ
AAnd χ
BObtain state space χ
AB(create " macroscopic view " failure state A
f^B
f, and remove all system's inaccessible states in the set);
2, " Macroscopic Process " parameter p
Ih, jlBy parameter p
IjAnd p
HlProduct be transformed, namely merge according to formula (1) and formula (2) state transitions with process A and process B.Wherein, p
IjBe the probability that state i transforms to state j, p
HlBe the probability that state h transforms to state l, p
Ih, jlFor having simultaneously state i and state h to the probability that has simultaneously state j and state l conversion.
(3) modeling process of " behavioural analysis module " in the mixed fault Early-warning Model:
" behavioural analysis module ", " functional analysis module " and the output of " degradation analysis module " are merged, and adopt further the DBN parameter estimation algorithm that the parameter of mixed fault Early-warning Model is quantitatively estimated, thereby set up the more accurate quantitative mixed fault Early-warning Model take DBN as framework, this model can quantitatively calculate and provide the concrete numerical value of estimating and predicting, rather than analyze qualitatively and describe, concrete syncretizing mechanism is as follows:
Theoretical based on dynamic bayesian network, each degenerative process that " degradation analysis module " exported is converted to mixed fault Early-warning Model dynamic variable node; Each transport stream attribute of " functional analysis module " output is converted to mixed fault Early-warning Model static variable node.Simultaneously, take the dynamic bayesian network Directed Graph Model as framework, according to the cause-effect relationship of " functional analysis module " output, (the father node) is done its direct result node (child node) of an oriented arrow points, until all nodes all are connected on the DBN network from the reason node.Namely, system architecture information (being described by assembly and transport stream), qualitative cause-effect relationship and dynamic degenerative process are integrated, the system failure chain of causation is carried out quantification modeling based on DBN, generate that the behavioural analysis module thisly comprises dynamically, the network of the oriented line between static node and each node is the structure of mixing Early-warning Model.And the quantitative relationship of line between variable information corresponding to each node (such as the implicit state variable of each node and conditional probability, priori failure probability etc.) and the node (such as, the transition probability of variable etc. between the joint probability of a group node, the different time fragment) is defined as the parameter of mixing Early-warning Model.These parameters can be estimated by the sample statistics learning method based on a large amount of historical datas.
Wherein, related algorithm those skilled in the art of above-mentioned mixing Early-warning Model parameter learning can consult following list of references: Baum L E, Petrie G S, Weiss N.A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains.The Annals of Mathematical Statistics, 1970,41 (1): 164 – 171.
(4) modeling process of " event analysis module " in the mixed fault Early-warning Model:
The purpose that generates " event analysis module " is, be provided at the latest data of the parameter (implicit state variable and observed parameter variable etc.) that defines in the above-mentioned mixed fault Early-warning Model, namely the node variable in the mixed fault Early-warning Model is dynamically updated, and provide the reasoning foundation for the fault pre-alarming analysis.Therefore, should " event analysis module " provide following two kinds of information for the mixed fault Early-warning Model:
1. the real-time monitored parametric variable value that provides condition monitoring system to gather, namely the data value (such as force value, flow value etc.) of observed parameter variable corresponding to mixed fault Early-warning Model static node periodically updated that (the concrete cycle is looked condition monitoring system, decide such as the data acquiring frequency of SCADA system), claim " monitoring event analysis submodule ";
2. for the supporting assembly in each process, based on various alternative maintenance schedules (causing the action in the future that mixed fault Early-warning Model node state changes), the update condition of implicit state variable corresponding to mixed fault Early-warning Model dynamic node and the data of renewal are provided, claim " maintenance event analysis submodule ".
(5) the early warning analysis process of mixed fault Early-warning Model " evaluation module ":
The function of " evaluation module " is to utilize DBN filtering inference mechanism by the up-to-date evidence that " event analysis module " provides, determine the variate-value of each node in the mixed fault Early-warning Model, especially implicit state variable value, and and then obtain security control measure and the detection scheme of possible consequence and the correspondence of the root sexual factor of fault, failure cause at different levels, fault before and after (" Spatial Dimension ") carries out in the same timeslice of the behavioural analysis module of mixed fault Early-warning Model to reasoning.
In the implementation of whole evaluation module, constantly follow the trail of the currency of implicit state variable, i.e. known observed parameter variable Y
tValue (being provided by " event analysis module ") uninterruptedly estimates implicit state variable X by following filtering reasoning algorithm
t:
P(X
t|y
1:t)∝P(y
t|X
t,y
1:t-1)P(X
t|y
1:t-1)＝P(y
t|X
t)[Σx
t-1P(X
t|x
t-1)P(x
t-1|y
1:t-1)] （3）
(6) the early warning analysis process of mixed fault Early-warning Model " prediction module ":
The function of " prediction module " is based on the implicit state variable value of parts that mixed fault Early-warning Model " evaluation module " provides, the Observable state variable Real-Time Monitoring value that " event analysis module " provides, by DBN Forecast reasoning mechanism, (" time dimension ") carries out reasoning between the different timeslice of the behavioural analysis module of mixed fault Early-warning Model, obtain implicit state variable following (the various probability that may values of τ＜t), thus realize the fiduciary level development trend (take the implicit state probability of prediction as index) in prognoses system or parts futures.
The Forecast reasoning mechanism of " prediction module " is carried out the probability that one-step prediction calculates the following energy of implicit state variable value
Or the probability of the following energy of calculating observation parametric variable value
Formula as follows:
Or
α wherein
t(i)=P (X
t=i|y
1:t), representing known observed parameter variable y when all values of time period [1, t], implicit state variable X is the probability of state i in moment t value; P (x
T+1| x
t) when representing known arbitrary variable current time state, the probability of following next constantly state generation.
The Output rusults of " prediction module " can be used for maintenance decision: can " safety " or " success " completion system set objective if the predicted value of to-be is met safety standard and can be considered, and then do not take any maintenance or detect action.Otherwise must carry out immediately (based on security consideration) maintenance task, or (preventative maintenance) carried out in a plan optimal time.
Therefore, execution by a series of " functional analysis modules ", " degradation analysis module ", " behavioural analysis module ", " event analysis module ", and it is merged the device systems fault chain of causation is carried out the quantification modeling, thereby made up the mixed fault Early-warning Model, this is the modelling phase of mixed fault Early-warning Model.In the fault pre-alarming analysis phase, this model carries out reasoning at " Spatial Dimension " and " time dimension " respectively by " evaluation module " and " prediction module ", realization is to the tracking of system failure root sexual factor, the reasoning of fault possibility consequence, the prediction of system degradation trend, and for the preventative maintenance the establishment of the project provides foundation, thereby essential safety that can safeguards system.
Below come the such scheme of the further description embodiment of the invention by the example of reality.
" puckery-Ning-Lan " gas transmission line east of a lake boosting station rock gas Compressor Set system (GTCS) is comprised of Suo La god of unusual strength's 130 type gas turbine and graceful turbine RV050/04 type centrifugal compressor.Because it forms complex structure, regulating working conditions is frequent, physical environment is abominable; in order to improve equipment essential safety level; avoid occuring the accidents such as disorderly closedown, natural gas leaking; must carry out modeling to the fault chain of causation of Compressor Set system; dynamic evaluation, predict its safe condition, and its remaining life is assessed.
(Compressor Set system and key part structure are seen shown in accompanying drawing 3a-Fig. 3 d) such as Condition Monitoring Data, historical maintenance record, running status daily sheet and dependent failure statistical data of investigation and collection Compressor Set system, obtain sufficient priori and expertise about Compressor Set system and similar rotating machinery, provide rational foundation to the correction that mixes Early-warning Model parameter and structure.Wherein, the critical component that illustrates of Fig. 3 d comprises: main oil pump, gas outlet, air motor and draft tube.
According to above-mentioned modeling method and the detailed example described in the embodiment of the invention, set up the mixed fault Early-warning Model for the Compressor Set system failure chain of causation, comprise the steps:
The first step is analyzed based on HAZOP and FMEA, generates " functional analysis module ", determines the main failure mode, process variable deviation of each key subsystem of Compressor Set system or parts, possible dangerous reasous and results of wrong subjects etc.
Particularly, it is common method in the system safety engineering subject that HAZOP and FMEA analyze, and those skilled in the art can be with reference to following document (wherein having detailed possible dangerous reason and the analytical procedure of consequence): [1] Shao Hui, " system safety engineering ", petroleum engineering publishing house, 2008.5.
Second step, FMEA analysis result and discrete time Markov stochastic process theory based on the first step, generate " degradation analysis module ", utilize the existing historical fail data of Compressor Set, fault data, equipment operation statistics and on-the-spot expertise, for the failure mode of subsystems or critical component is set up quantitative degenerative process, and determine its parameter.
The 3rd step under the dynamic bayesian network framework, generated " behavioural analysis module ", and " functional analysis module " and " degradation analysis module " acquired results are merged, and set up the mixed fault Early-warning Model.Wherein with corresponding dynamic node and the correlation parameter that generates the mixed fault Early-warning Model of the degenerative process of " degradation analysis module " gained unit state.This correlation parameter comprises transition probability and the priori failure probability of state space.Corresponding static node and the correlation parameter that generates the mixed fault Early-warning Model of observed parameter variable with " functional analysis module " gained unit.Said units refers to the set of subsystem, parts after the system decomposition etc.Simultaneously in the mixed fault Early-warning Model father node and child node set and between line shown the network structure of mixed fault Early-warning Model, as shown in Figure 5, i.e. the reciprocation of state evolution and parameter variation between the unit in the characterization system.In Fig. 5, D5_1: oil pipe degenerative process (OTD); D5_2: oil cooler degenerative process (OCD); D5_3: oily filter element part degenerative process (OFD); D5_4: oil pump degenerative process (OPD); D5_5: lubricating oil heater degenerative process (OHD); S5_1: charge oil pressure; S5_2: oil temperature supplying: S5_3: oil tank liquid level: S5_4: oil filter pressure reduction; S5_5: fuel tank temperature; E1_1: environment temperature.White circle represents dynamic node among Fig. 5; Gray circles represents static node; Wherein, K represents the current time unit, K-1 representative upper a time quantum constantly.
It is to set up on the basis of existing data (2005-2009 year machine operation, and part equipment maintenance records before in 2005 etc.) that Compressor Set mixes parameters in the Early-warning Model.But because the continuous variation of system running environment must provide a real-time or regular update mechanism to above-mentioned model parameter, so that the result of mixed fault Early-warning Model meets the demand of current safety management and operation.Therefore, to the Compressor Set system made " event analysis module ", its software interface is (software is based on C++Builder6.0 and the SQL Server2005 platform and develops) as shown in Figure 4.Compressor Set " event analysis module " application software is utilized OPC(OLE for Process Control; the OLE that is used for process control) the remote high-speed data acquisition technology is carried out data acquisition and status monitoring; and various real time datas are classified (such as start/stop machine data, regulating working conditions data, normal service data, fault state data etc.); store in the safety database relevant with the monitoring event in " event analysis module ", be convenient to data retrieval and processing." event analysis module " can also regularly store equipment maintenance record in the safety database relevant with maintenance event in " event analysis module " into, is convenient to regularly fail data be carried out statistical study.Along with " event analysis module " constantly gathers and store various valid data, dynamic correction and renewal for the parameter in the mixed fault Early-warning Model and network structure provides rational foundation on the one hand, makes model more accurately also can meet the actual conditions of environment, operating mode operation adjusting; Reasoning process for " evaluation module " and " prediction module " in the mixed fault Early-warning Model (seeing for fourth, fifth step) provides soft/hard evidence on the other hand, strengthens fault rootstock sexual factor identification and failure trend prediction result's rationality.
In the 4th step, mixed fault Early-warning Model " evaluation module " is verified system failure root sexual factor identification process.
Carry out real-time online status monitoring to Qinghai Lake eastern station rock gas Compressor Set in May, 2008, it is 57 ° of C that certain day records the lubricating oil medial temperature, lubricating oil pressure is 0.195MPa, (the lubricating oil normal temperature is interval for [35 all to depart from range of normal value, 55] ° C, the normal span of lubricating oil pressure is [0.210,0.449] MPa).The dangerous reason that tradition HAZOP analysis report is possible with 10 shown in the indicator gauge 1 even redundant analysis result occurs, and each dangerous reason exports side by side regardless of primary and secondary, brought difficulty to security decision.If the model node surpasses 10, then traditional HAZOP analysis result will be very huge, " shot array " situation often occur.Security of system person checks one by one according to tabulation, i.e. wasting manpower and material resources, and the system that may miss again best servicing time, thus system running state is worsened.The fault that the embodiment of the invention provides is mixed Early-warning Model and then be can be good at solving this difficult problem.
Now according to " evaluation module " reasoning algorithm of describing in the embodiment of the invention, each node state probability calculation analytic process is as follows:
(1) the observer nodes deviation is blured quantification as node posterior probability in the mixed fault Early-warning Model.Wherein, about the technological means of fuzzy quantification, those skilled in the art can be with reference to such as Publication about Document: Hu Jinqiu, Zhang Laibin, Liang Wei, imperial court's sunshine. quantitative HAZOP analyzes based on the Compressor Set of Fuzzy Data Fusion. the system engineering theory and practice, 2009,29(8): 153-159.Lubricating oil normal temperature interval is [35,55] ° C during unit operation, and now recording the lubricating oil medial temperature is 57 ° of C, and then the posterior probability of observer nodes S7_2 gets { 0 as calculated, 0.1546,0.7939,0.0515 }, it is { on the low side to correspond respectively to possible state, normally, higher, superelevation }.In like manner the normal span of lubricating oil pressure is [0.210,0.449] MPa, and then the posterior probability of observer nodes S7_1 gets { 0.0593,0.8536,0.0871,0 } as calculated, corresponds respectively to possible state { ultralow, on the low side, normal, higher }.Wherein, posterior probability is to process according to fuzzy quantification to calculate, for example, each possible state (on the low side, higher, normal etc.) corresponding fuzzy membership function separately, then actual observed value is brought into, calculated the fuzzy membership of each state, as the posterior probability of each state.Unusual observed reading all occurs in the oil system, and the network structure of its mixed fault Early-warning Model as shown in Figure 5.The dynamic node D5_1 corresponding with system component, D5_2, D5_3, D5_4, D5_5 are the reliablity estimation value that the probability of normal condition can be used as these parts, through " evaluation module " reasoning, it is respectively 0.3140,0.6583, and 0.7244,0.5395 and 0.9016.Wherein to be positioned at the probable value of normal condition minimum for D5_1 and D5_4 node, shows that the reliability of lube pipe and oil pump is relatively poor, has potential safety hazard, needs to check or repair.
Table 1: " evaluation module " the reasoning results in the mixed fault Early-warning Model-possible dangerous reason and safety practice
(2) the possible dangerous reasous and results of wrong subjects (fault rootstock sexual factor) that is existed to the inference estimation oil system before and after " Spatial Dimension " by " evaluation module " of mixed fault Early-warning Model sees Table 1 with shown in the table 2.Its demonstration has the dangerous former of maximum possibility occurrence (0.8035) because " the meter in aperture is less than normal "; Secondly be " the outlet water temperature is higher ", possibility is 0.7142; Dangerous consequences with maximum possibility occurrence (0.9132) for " pressure film is difficult to keep, in addition cause that bearing shell burns out, the accidents such as sealing wear and impeller destruction ".The device security maintainer presses from big to small successively maintenance and inspection of possibility occurrence according to the dangerous reason table of system and corresponding suggestion safety practice.The actual inspection result conforms to " evaluation model " the reasoning results, the potential safety hazard that the operation of discovery system is potential.Simultaneously formulate the security control measure according to the dangerous consequences of maximum likelihood, the associated arguments such as the temperature of bearing shell, sealing, displacement are carried out close supervision.The on-site maintenance personnel fall into the normal value interval by increasing cooling water inflow envoy point observation value, have avoided the continuation of fault to worsen; And the observed reading that notes abnormalities organize two days later the Compressor Set clean-out operation, emphasis cleans the throttle orifice plate, and maintenance lubricating system oil pump, oil pipe, cleans oil filter.
Table 2: the reasoning results of mixed fault Early-warning Model " evaluation module "-possible dangerous consequences
Wait safety evaluation, Analysis of Fault Diagnosis to compare with the single HAZOP analysis of tradition, the mixed fault Early-warning Model that the embodiment of the invention provides is conducive to find fast and accurately the fault rootstock sexual factor and formulates safety practice, solve the limitation of traditional HAZOP qualitative analysis with uncertain, can not export huge and redundant the reasoning results.The mixed fault Early-warning Model has strengthened and carried out the accuracy of identification and the rationality of decision-making when there are a plurality of node deviations (being a plurality of dangerous matter sources) in system simultaneously.The on-site maintenance personnel only need to check successively according to failure cause possibility occurrence size, often check that 2-3 factor can locate fault rootstock, thereby initiatively take in advance maintenance measures with the factor controlling of these induced malfunctions in a rational intensity or level, to prevent from bringing out the further trouble or failure of parts.
In the 5th step, mixed fault Early-warning Model " prediction module " is to system unit degradation trend and the checking of remaining life forecasting process.
According to above-mentioned " prediction module " reasoning algorithm in the embodiment of the invention, go up Inference Forecast oil system reliability development trend in future from current degenerate state at " time dimension ", as shown in Figure 6.As seen the D5_1(oil pipe that current fiduciary level is minimum) and the D5_4(oil pump) to use its degree of degeneration of growth at age with parts will be more serious, and the fiduciary level of its prediction after 30 days will be lower than 0.5.If fiduciary level 50% threshold value commonly used on the engineering is declared the abandoned tender standard as equipment failure, then its remaining life is predicted as 30 days.Simultaneously, although the D5_3(oil strainer) current fiduciary level is positioned at security level, and its deterioration velocity is very high, if untimely cleaning or replacing will affect the normal operation of miscellaneous part.Consider the relevant failure problem, the oil system of degeneration on the impact of turbine system centre bearer as shown in Figure 7.It shows at lubricating oil temperature higher, in the situation that lubrication pressure is on the low side, and bearing (D5_1:GT1
#2
#3
#Bearing; The D5_2:GT thrust bearing) degradation trend (solid line D5_1 and D5_2 among Fig. 7) is more serious than normal degradation trend (corresponding respectively to dotted line D5_1 ' and D5_2 ' among Fig. 7), and namely the degenerate state of oil system will shorten the serviceable life of bearing arrangement.
The evaluation that provides according to the mixed fault Early-warning Model of the embodiment of the invention with predict the outcome, formulate maintenance program, and maintenance operation quantitatively turned to the alternative maintenance schedule in " event analysis module ", and implicit state corresponding to the dynamic node in the mixed fault Early-warning Model exerted an influence, detailed process is as follows:
(1) renewal of oil pipeline, and this operation is considered as keeping in repair fully, the state that then keeps in repair posterior nodal point D5_1 is updated to [1 00 0];
(2) clean oil filter, and this operation is considered as imperfect repair, the state that then keeps in repair posterior nodal point D5_3 is updated to [0.9 0.1 0];
(3) keep in repair oil pump, and this operation is considered as imperfect repair, the state that then keeps in repair posterior nodal point D5_4 is updated to [0.85 0.15 0].
Fig. 8 has provided after " event analysis module " revised the mixed fault Early-warning Model, namely carries out the development trend of the following oil system fiduciary level of Forecast reasoning from the state that keeps in repair rear renewal.As seen in 60 days of future, all parts serviceability is good, and fiduciary level substantially all keeps (being considered as more than 0.5 and can accepting according to the engineering experience fiduciary level, and be in comparatively safe state) more than 0.5.Because the degradation mechanism of oil strainer self and the effect of external environment condition, the deterioration velocity of the parts that the D5_3 node is corresponding (oil strainer) is compared with miscellaneous part comparatively fast, therefore maintenance (be generally and clean or the change) cycle of oil strainer can be formulated and is every 60-90 days, thereby guarantee the normal operation of system's miscellaneous part.
The 6th step, the formulation of preventative maintenance strategy.
The evaluation of mixed fault Early-warning Model can be used as the input of maintenance decision link with predicting the outcome: when system's (parts) reliability is lower than a certain setting value (or standard), to in " event analysis module ", trigger corresponding maintenance action, thereby the relevant dynamic node state of mixed fault Early-warning Model will be upgraded.Take the Compressor Set air system as example, suppose wherein two parts (D1_1: air strainer and D1_2: ventilating system) all be in brand-new state, in the system degradation trend following in 300 days shown in Fig. 9 a.When t=168(days) time, the fiduciary level of air strainer (D1_1) drops to 0.5(namely, and the air strainer remaining life is predicted as 168 days), and ventilating system (D1_2) work is good, without obvious degradation trend.Analyze the observer nodes (S1_1 of representative system operation conditions, S1_2, S1_3, S1_4) (wherein S1_1 is that inlet filter pressure reduction, S1_2 are that tank pressure, S1_3 are that spin manifold temperature, S1_4 are to development trend: the gas turbine inlet air temperature), shown in Fig. 9 b, S1_1(inlet filter pressure reduction) will depart from gradually normal value, and will be to its miscellaneous part that is associated have a negative impact (such as compressor system, combustion system etc.).
In this case, the maximum preventative maintenance cycle of air strainer can be set as 168(days) (corresponding to 50% safety reliability standard).If guarantee sufficient security of system surplus, maintenance cycle can be set as 60-90(days) (when fiduciary level is lower than 0.7).Figure 10 a has shown that the air system under 60 days maintenance cycle conditions degenerates and the operating index development trend.Wherein maintenance action is assumed to the cleaning to air strainer, and be considered as imperfect repair (generally because the degeneration of component materials and have certain attrition and attack, parts can't reach brand-new state after cleaning), then air strainer D1_1 unit status is updated to [0.9 0.1 0] after the maintenance.
Shown in Figure 10 b, in the presence of maintenance schedule, model prediction goes out the fiduciary level of air entire system all more than 0.7, and the probability that observer nodes is in normal condition also is more than 0.7, and it shows that the formulation of this maintenance schedule makes air system can keep for a long time good safe work state.Simultaneously, although predicting the outcome, this has also shown the regular preventative maintenance of having considered air system, but still there is certain degradation trend in observational variable S1_2, its reason is that S1_2 is subjected to the impact of air strainer and ventilating system simultaneously, and when time span was longer, ventilating system also exists to degenerate slowly and affect node S1_2 made it depart from gradually normal value.
Present embodiment is by the rig-site utilization example to the Compressor Set system, set forth the foundation of mixed fault Early-warning Model of the present invention and definite overall process of model parameter, and given sufficient checking according to the collection in worksite data to correctness, validity and the rationality of model pre-warning analysis.
The beneficial effect of the embodiment of the invention is:
1, in order to strengthen the rationality of fault rootstock sexual factor identification, the mixed fault Early-warning Model in the embodiment of the invention has following outstanding characteristics when processing complication system state evolution uncertain problem:
(1) can process uncertain and probabilistic event of failure;
(2) can be used in the study Failure causality;
(3) be a kind of (expert) priori, measured data and physical mechanism to be carried out comprehensive better expression pattern;
(4) can process the data set of imperfect information (or partial data is lost);
(5) dirigibility is better, can continue the performance with expertise and data improved model, and model structure and parameter have clear and definite implication.
2, pass through quantitative modeling and the reasoning of the fault chain of causation, the required assumed conditions such as component failure independence of classic method have been broken through, effectively realized fault pre-alarming analyze in to the accurate identification of multi-part, many dangerous matter sources system failure root sexual factor, and in the fault pre-alarming control to the reasonable prediction of remaining life:
(1) " functional analysis module " set up the relational model that influences each other (reciprocal process) between the system unit and between the fault by functional analysis and failure analysis in the mixed fault Early-warning Model modeling method of the present invention, how to have solved the difficult problem of qualitative examination device systems Failure causality.
(2) " degradation analysis module " by to the degeneration physical mechanism of influential system supporting assembly carry out statistical modeling, solved a difficult problem that how to represent fault chain of causation node in the fault pre-alarming model.
(3) " behavioural analysis module " carries out system's " functional analysis module " comprehensively with " degradation analysis module " acquired results, digraph take dynamic bayesian network as the system failure chain of causation is expressed, and has solved the difficult problem how quantification makes up the fault chain of causation.
(4) " event analysis module " is by real-time follow-up monitoring event and maintenance event, proposed a kind of mixed fault Early-warning Model (structure and parameter) dynamically to be revised and the mechanism of upgrading, solved because the adjusting of the variation of external environment condition, operating mode operation and maintenance activity causes the unconformable difficult problem of Early-warning Model.
(5) evidence that provides by " event analysis module " of " evaluation module " in the mixed fault Early-warning Model, utilize DBN filtering inference mechanism, draw the probability of the implicit state that each dynamic node is corresponding the mixed fault Early-warning Model from " Spatial Dimension " reasoning, solved how based on the root sexual factor of fault chain of causation reasoning identification of defective, possible consequence and the corresponding Pre-control measures difficult problem of fault, the condition of having avoided fault to produce.
(6) " prediction module " in the mixed fault Early-warning Model be by DBN Forecast reasoning mechanism, from the degeneration evolution process in " time dimension " prognoses system or parts futures, how to have solved the development trend difficult problem based on fault chain of causation Inference Forecast fault.
The above; only for the better embodiment of the present invention, but protection scope of the present invention is not limited to this, anyly is familiar with those skilled in the art in the technical scope that the present invention discloses; the variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.
Claims (8)
1. the modeling method of a mixed fault Early-warning Model is characterized in that, described method comprises:
Analyze HAZOP or failure mode and effect analysis FMEA, systematic function analysis module based on hazard and operability; Described functional analysis module is used for determining dynamic node and the static node of mixed fault Early-warning Model, the implicit state variable that dynamic node is corresponding and the state space of implicit state variable, the observational variable that static node is corresponding and the state space of observational variable, and, reach the reciprocal effect relation between each parts and the environment between all parts; Wherein, described functional analysis module is that employing two-way function failure analysis mechanism is set up the relational model that influences each other between the system unit;
Based on analysis result and the discrete time Markov theory of random processes of FMEA, generate the degradation analysis module; Described degradation analysis module is used for component failure data and priori according to historical data base, determines state transitions rule and the failure probability density function of the implicit state variable that each dynamic node of mixed fault Early-warning Model is corresponding;
According to Condition Monitoring Data and maintenance information, generate event analysis module; Described event analysis module is used for real time data and the chronic frustration data that monitor are stored into described historical data base, for the reasoning process of evaluation module and prediction module provides the reasoning evidence; Wherein, described Condition Monitoring Data comprises that described maintenance information comprises maintenance record or various preventive maintenance schedule from Real-time Monitoring Data and the various Action Events record of spot sensor, Programmable Logic Controller or data acquisition and supervisor control acquisition;
According to the Output rusults of described functional analysis module and described degradation analysis module, theoretical in conjunction with dynamic bayesian network DBN, generate the behavioural analysis module; Described behavioural analysis module, be used for setting up network structure and the parameter of mixed fault Early-warning Model, utilize simultaneously Condition Monitoring Data, the fail data in the described historical data base and adopt the DBN parameter estimation algorithm parameter of described mixed fault Early-warning Model is estimated and to be upgraded;
Take the Observable state variable Real-Time Monitoring value of described event analysis module output as the reasoning evidence, utilize the dynamic bayesian network reasoning algorithm in the same timeslice of described behavioural analysis module, to carry out front and back to reasoning, generate evaluation module, with the possible consequence of output system failure factor and fault initiation; Described evaluation module, be used for utilizing the dynamic bayesian network reasoning algorithm in the same timeslice of described behavioural analysis module, to carry out front and back to reasoning, take the Observable state variable Real-Time Monitoring value of described event analysis module output as the reasoning evidence, the implicit state variable that each dynamic node of derivation mixed fault Early-warning Model is corresponding from the Spatial Dimension and corresponding probability of happening, thereby to the implicit state of the current all parts of terminal output system, the fault rootstock sexual factor, dangerous reasons at different levels, possible dangerous consequences, in safety practice and the turnaround plan at least one; Described evaluation module also is used for constantly following the trail of the currency of implicit state variable, and the currency of described implicit state variable is known observed parameter variable Y
tValue uninterruptedly estimates implicit state variable X by following filtering reasoning algorithm
t: P (X
t| y
1:t) ∝ P (y
t| X
t, y
1:t-1) P (X
t| y
1:t-1)=P (y
t| X
t) [Σ x
T-1P (X
t| x
T-1) P (x
T-1| y
1:t-1)]; Wherein, y
1:tObserved parameter variable y the time period 1,2 ..., t-1, the sequence of observations on the t}, y
tThat observed parameter variable y is at t observed reading constantly, y
1:t-1Observed parameter variable y the time period 1,2 ..., t-2, the sequence of observations on the t-1}, x
T-1That implicit state variable x is at t-1 state value constantly, X
tThat implicit state variable x is in t state estimation value constantly;
Take the Observable state variable Real-Time Monitoring value of the Output rusults of described evaluation module and the output of described event analysis module as the reasoning evidence, that utilizes the described behavioural analysis module of dynamic bayesian network reasoning algorithm carries out front and back to reasoning between different timeslices, the generation forecast module is with the output system all parts degradation trend in future; Described prediction module is used at least one according to each observed parameter of the implicit State-output system of described all parts variate-value in future, the parts degradation trend in future, remaining life, predictive maintenance strategy.
2. method according to claim 1 is characterized in that, described analysis result and discrete time Markov theory of random processes based on FMEA generate the degradation analysis module and comprise:
From the output of described functional analysis module, select crucial degeneration/failure mode to set up degenerative process;
Correlativity and interaction between the described degenerative process of identification, wherein said interaction comprise between the different degenerative processes of same parts and the interaction between the degenerative process of different parts.
3. method according to claim 2 is characterized in that, selects crucial degeneration/failure mode to set up degenerative process from the output of described functional analysis module and comprises:
From the output of described functional analysis module, select crucial degeneration/failure mode, and utilize discrete time markov stochastic process DTMP to set up degenerative process.
4. method according to claim 1 is characterized in that, and is theoretical in conjunction with dynamic bayesian network according to the Output rusults of functional analysis module and degradation analysis module, generates the behavioural analysis module and comprises:
Based on dynamic bayesian network, each degenerative process in the described degradation analysis module is converted to the dynamic variable node of mixed fault Early-warning Model;
Each Observable parameter in the described functional analysis module is converted to the static variable node of mixed fault Early-warning Model;
Based on dynamic bayesian network, the cause-effect relationship according between the variable of described functional analysis module output connects described dynamic variable node, described static variable node, forms the fault chain of causation.
5. method according to claim 1, it is characterized in that, take the observable variable Real-Time Monitoring value of described event analysis module output as the reasoning evidence, utilize the dynamic bayesian network reasoning algorithm in the same timeslice of described behavioural analysis module, to carry out front and back to reasoning, form evaluation module and comprise:
According to the reasoning evidence that described event analysis module provides, utilize the estimation inference mechanism of dynamic bayesian network, determine the implicit state probability of each dynamic node variable in the mixed fault Early-warning Model;
Obtain the factor of fault, consequence and the corresponding security control measure of fault to reasoning before and after in the same timeslice of described behavioural analysis module, carrying out.
6. method according to claim 1, it is characterized in that, take the Observable state variable Real-Time Monitoring value of described evaluation module Output rusults and event analysis module output as the reasoning evidence, utilize the dynamic bayesian network reasoning algorithm between the different timeslice of described behavioural analysis module, to carry out front and back to reasoning, form prediction module and comprise:
The reasoning evidence that provides according to described event analysis module, and take the current implicit state probability of the dynamic node variable of evaluation module output as the basis, utilize dynamic bayesian network Forecast reasoning mechanism, determine the implicit probability distribution over states on future a series of time points of each dynamic node variable in the described fault pre-alarming model;
To reasoning, obtain all parts degradation trend in future before and after between the different time sheet of described behavioural analysis module, carrying out, by setting failure threshold, the remaining life of further determining means.
7. the modeling of a mixed fault Early-warning Model is characterized in that, comprising:
The functional analysis module, be used for determining dynamic node and the static node of mixed fault Early-warning Model, the implicit state variable that dynamic node is corresponding and the state space of implicit state variable, the observational variable that static node is corresponding and the state space of observational variable and, between all parts and the relation of the reciprocal effect between each parts and the environment; Wherein, described functional analysis module is that employing two-way function failure analysis mechanism is set up the relational model that influences each other between the system unit;
The degradation analysis module is used for component failure data and priori according to historical data base, determines state transitions rule and the failure probability density function of the implicit state variable that each dynamic node of mixed fault Early-warning Model is corresponding;
Event analysis module, described event analysis module is according to Condition Monitoring Data and maintenance Information generation, be used for real time data and the chronic frustration data that monitor are stored into described historical data base, for the reasoning process of evaluation module and prediction module provides the reasoning evidence; Wherein, described Condition Monitoring Data comprises that described maintenance information comprises maintenance record or various preventive maintenance schedule from Real-time Monitoring Data and the various Action Events record of spot sensor, Programmable Logic Controller or data acquisition and supervisor control acquisition;
The behavioural analysis module, be used for setting up network structure and the parameter of mixed fault Early-warning Model, utilize simultaneously Condition Monitoring Data, the fail data in the described historical data base and adopt dynamic bayesian network DBN parameter estimation algorithm the parameter of described mixed fault Early-warning Model is estimated and to be upgraded;
Evaluation module is used for implicit state, system failure root sexual factor, dangerous reason at different levels, dangerous consequences and the corresponding safety practice through reasoning output current system all parts, at least one of turnaround plan; Described evaluation module, concrete being used for utilizes the dynamic bayesian network reasoning algorithm to carry out front and back to reasoning in the same timeslice of described behavioural analysis module, take the Observable state variable Real-Time Monitoring value of described event analysis module output as the reasoning evidence, the implicit state variable that each dynamic node of derivation mixed fault Early-warning Model is corresponding from the Spatial Dimension and corresponding probability of happening, thereby to the implicit state of the current all parts of terminal output system, the fault rootstock sexual factor, dangerous reasons at different levels, possible dangerous consequences, in safety practice and the turnaround plan at least one; Described evaluation module also is used for constantly following the trail of the currency of implicit state variable, and the currency of described implicit state variable is known observed parameter variable Y
tValue uninterruptedly estimates implicit state variable X by following filtering reasoning algorithm
t:
P (X
t| y
1:t) ∝ P (y
t| X
t, y
1:t-1) P (X
t| y
1:t-1)=P (y
t| X
t) [Σ x
T-1P (X
t| x
T-1) P (x
T-1| y
1:t-1)]; Wherein, y
1:tObserved parameter variable y the time period 1,2 ..., t-1, the sequence of observations on the t}, y
tThat observed parameter variable y is at t observed reading constantly, y
1:t-1Observed parameter variable y the time period 1,2 ..., t-2, the sequence of observations on the t-1}, x
T-1That implicit state variable x is at t-1 state value constantly, X
tThat implicit state variable x is in t state estimation value constantly;
Prediction module is used at least one according to each observed parameter of the implicit State-output system of described all parts variate-value in future, the parts degradation trend in future, remaining life, predictive maintenance strategy.
8. the modeling of mixed fault Early-warning Model according to claim 7 is characterized in that, described event analysis module comprises:
Monitoring event analysis submodule is used for the data value of observed parameter variable corresponding to mixed fault Early-warning Model static node is periodically updated;
Maintenance event is analyzed submodule, is used for according to alternative maintenance schedule, and the update condition of implicit state variable corresponding to mixed fault Early-warning Model dynamic node and the data of renewal are provided.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110150703 CN102262690B (en) | 2011-06-07 | 2011-06-07 | Modeling method of early warning model of mixed failures and early warning model of mixed failures |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110150703 CN102262690B (en) | 2011-06-07 | 2011-06-07 | Modeling method of early warning model of mixed failures and early warning model of mixed failures |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102262690A CN102262690A (en) | 2011-11-30 |
| CN102262690B true CN102262690B (en) | 2013-10-30 |
Family
ID=45009316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110150703 Expired - Fee Related CN102262690B (en) | 2011-06-07 | 2011-06-07 | Modeling method of early warning model of mixed failures and early warning model of mixed failures |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102262690B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104696207A (en) * | 2013-12-04 | 2015-06-10 | 铜陵市永生机电制造有限责任公司 | Method for identifying dangerousness of compressor unit |
Families Citing this family (49)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102567639A (en) * | 2011-12-30 | 2012-07-11 | 南京航空航天大学 | Method for evaluating reliability of aircraft engine aiming at competing failure |
| CN104350435B (en) * | 2012-04-13 | 2016-12-14 | 西门子公司 | The prediction analysis method being embedded in the software platform of programmable logic controller (PLC) |
| JP5751496B2 (en) * | 2012-12-27 | 2015-07-22 | 横河電機株式会社 | Event analysis apparatus and computer program |
| CN104102804A (en) * | 2013-04-11 | 2014-10-15 | 华为技术有限公司 | Method and device for predicting service life of component of equipment |
| JP5769138B2 (en) | 2013-04-22 | 2015-08-26 | 横河電機株式会社 | Event analysis apparatus and computer program |
| WO2015077890A1 (en) * | 2013-11-27 | 2015-06-04 | Adept Ai Systems Inc. | Method and system for artificially intelligent model-based control of dynamic processes using probabilistic agents |
| WO2015179604A1 (en) * | 2014-05-21 | 2015-11-26 | Cooper Technologies Company | Enclosure diagnostic and control systems |
| CN104238545B (en) * | 2014-07-10 | 2017-02-01 | 中国石油大学(北京) | Fault diagnosis and pre-warning system in oil refining production process and establishment method thereof |
| CN104504260B (en) * | 2014-12-16 | 2017-10-17 | 西北工业大学 | A kind of multi-invalidation mode system failure Forecasting Methodology |
| CN104850748B (en) * | 2015-05-26 | 2017-09-15 | 北京交通大学 | A kind of railway track fractures accident analysis method for early warning and system |
| CN105067295A (en) * | 2015-07-20 | 2015-11-18 | 柳州好顺科技有限公司 | Engineering machine fault alarm device |
| CN105184386A (en) * | 2015-07-22 | 2015-12-23 | 中国寰球工程公司 | Method for establishing abnormal event early warning system based on expert experience and historical data |
| CN105426970B (en) * | 2015-11-17 | 2018-02-13 | 武汉理工大学 | A kind of meteorological intimidation estimating method based on discrete dynamic Bayesian network |
| CN107133118B (en) * | 2016-02-26 | 2020-07-14 | 华为技术有限公司 | Fault diagnosis model training method, fault diagnosis method and related device |
| CN105930963B (en) * | 2016-04-15 | 2020-07-31 | 中国船舶工业系统工程研究院 | Electromechanical system equipment health assessment method |
| CN106021062B (en) * | 2016-05-06 | 2018-08-07 | 广东电网有限责任公司珠海供电局 | The prediction technique and system of relevant fault |
| EP3519967A4 (en) * | 2016-09-29 | 2020-06-03 | Hewlett-Packard Development Company, L.P. | Component failure prediction |
| CN106503813A (en) * | 2016-10-27 | 2017-03-15 | 清华大学 | Prospective maintenance decision-making technique and system based on hoisting equipment working condition |
| CN106779285A (en) * | 2016-11-16 | 2017-05-31 | 深圳市燃气集团股份有限公司 | A kind of failure mode analysis (FMA) implementation method of LNG storage tank leakage |
| CN106649569A (en) * | 2016-11-16 | 2017-05-10 | 上海建桥学院 | Data analysis and early-warning method based on information connotation degree |
| CN106600167B (en) * | 2016-12-29 | 2020-09-08 | 中国石油大学(华东) | Ocean platform fire risk assessment method considering human errors and tissue defects |
| CN107358299B (en) * | 2017-06-16 | 2022-11-22 | 杭州培慕科技有限公司 | Predictive maintenance closed-loop method based on fault mode |
| CN107301296B (en) * | 2017-06-27 | 2020-09-04 | 西安电子科技大学 | Data-based qualitative analysis method for circuit breaker fault influence factors |
| CN107294795A (en) * | 2017-08-02 | 2017-10-24 | 上海上讯信息技术股份有限公司 | A kind of network security situation prediction method and equipment |
| CN107562041A (en) * | 2017-09-22 | 2018-01-09 | 广东工业大学 | Goat fault early warning method, device, equipment and computer-readable recording medium |
| CN109697511B (en) * | 2017-10-24 | 2022-04-05 | 华为技术有限公司 | Data reasoning method and device and computer equipment |
| CN108959676B (en) * | 2017-12-22 | 2019-09-20 | 北京航空航天大学 | It is a kind of to consider the degeneration modeling effectively impacted and life-span prediction method |
| CN108255728B (en) * | 2018-01-18 | 2021-03-09 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Method and device for identifying failure mode of software |
| CN108491860B (en) * | 2018-02-24 | 2021-02-05 | 中国石油大学(北京) | Shale gas fracturing equipment full life cycle real-time reliability dynamic evaluation method and device |
| CN108197847B (en) * | 2018-03-22 | 2022-04-19 | 北京电子工程总体研究所 | Analysis system for recognizing fault hazard source of air-defense missile weapon system |
| CN110619133B (en) * | 2018-06-20 | 2023-10-13 | 大陆泰密克汽车系统（上海）有限公司 | Method for reducing single point failure rate in vehicle control system |
| CN109257206A (en) * | 2018-08-10 | 2019-01-22 | 南方电网科学研究院有限责任公司 | A kind of method of data inputting and evaluation diagnostic message feedback |
| CN109521299B (en) * | 2018-11-27 | 2020-11-03 | 河南理工大学 | Intelligent fault reasoning method for inverter |
| CN109657982B (en) * | 2018-12-20 | 2022-02-11 | 三一重能有限公司 | Fault early warning method and device |
| CN109858356B (en) * | 2018-12-27 | 2021-06-22 | 北京邮电大学 | Method and device for detecting input signal of unknown complex system |
| CN110162423B (en) * | 2019-05-21 | 2021-04-13 | 联想(北京)有限公司 | Resource checking method and resource checking device |
| CN110490218B (en) * | 2019-06-10 | 2022-11-29 | 内蒙古工业大学 | Rolling bearing fault self-learning method based on two-stage DBN |
| CN110489898B (en) * | 2019-08-26 | 2021-05-28 | 北京航空航天大学 | Dynamic multi-level system modeling and state prediction method based on hybrid cognition |
| CN111343016B (en) * | 2020-02-21 | 2021-01-26 | 北京京东尚科信息技术有限公司 | Cloud server cluster management method and device |
| CN111611718B (en) * | 2020-05-27 | 2023-04-07 | 江南大学 | Actuator fault estimation method based on Bayesian learning |
| CN112001545B (en) * | 2020-08-24 | 2022-03-15 | 中国石油大学(华东) | Digital twin-driven marine oil underwater production system fault prediction method and system |
| CN112115128A (en) * | 2020-09-15 | 2020-12-22 | 北京唯实兴邦科技有限公司 | 5G real-time dynamic fault processing system based on multi-dimensional sensor |
| CN112348071B (en) * | 2020-10-30 | 2022-05-17 | 浙江大学 | Open-circuit fault diagnosis method for switching tube of submodule of modular multilevel converter |
| CN113610315B (en) * | 2021-08-16 | 2023-10-13 | 傲林科技有限公司 | Event network-based prediction method and device and electronic equipment |
| CN113757223B (en) * | 2021-09-06 | 2023-11-03 | 江苏徐工工程机械研究院有限公司 | Hydraulic component reliability analysis method and system, computer device, and storage medium |
| CN114237203A (en) * | 2021-12-13 | 2022-03-25 | 中国船舶重工集团公司第七0三研究所 | Preventive maintenance implementation method for pneumatic control system of ship gas turbine based on virtual reality |
| US11694540B1 (en) | 2021-12-17 | 2023-07-04 | Honeywell International Inc. | Fire events pattern analysis and cross-building data analytics |
| CN116187982B (en) * | 2023-04-24 | 2023-08-11 | 成都盛锴科技有限公司 | Rail transit vehicle multi-source multi-mode fault risk reasoning and maintenance decision method |
| CN116629707B (en) * | 2023-07-20 | 2023-10-20 | 合肥喆塔科技有限公司 | FDC traceability analysis method based on distributed parallel computing and storage medium |
-
2011
- 2011-06-07 CN CN 201110150703 patent/CN102262690B/en not_active Expired - Fee Related
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104696207A (en) * | 2013-12-04 | 2015-06-10 | 铜陵市永生机电制造有限责任公司 | Method for identifying dangerousness of compressor unit |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102262690A (en) | 2011-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102262690B (en) | Modeling method of early warning model of mixed failures and early warning model of mixed failures | |
| Hu et al. | An integrated safety prognosis model for complex system based on dynamic Bayesian network and ant colony algorithm | |
| JP6159059B2 (en) | Plant operation optimization system and method | |
| Qiu et al. | Fault diagnosis of wind turbine with SCADA alarms based multidimensional information processing method | |
| US8751421B2 (en) | Machine learning for power grid | |
| CN104238545B (en) | Fault diagnosis and pre-warning system in oil refining production process and establishment method thereof | |
| Hu et al. | An integrated method for safety pre-warning of complex system | |
| CN106161138A (en) | A kind of intelligence automatic gauge method and device | |
| Roemer et al. | An overview of selected prognostic technologies with application to engine health management | |
| Vu et al. | A stationary grouping maintenance strategy using mean residual life and the birnbaum importance measure for complex structures | |
| Ciarapica et al. | Managing the condition-based maintenance of a combined-cycle power plant: an approach using soft computing techniques | |
| Hu et al. | DBN based failure prognosis method considering the response of protective layers for the complex industrial systems | |
| Vafaei et al. | Fuzzy early warning systems for condition based maintenance | |
| Chin et al. | Asset maintenance optimisation approaches in the chemical and process industries–A review | |
| CN105096053A (en) | Health management decision-making method suitable for complex process system | |
| CN104573845A (en) | Auxiliary decision-making method for equipment state maintenance of information system | |
| CN104392752A (en) | Real-time on-line nuclear reactor fault diagnosis and monitoring system | |
| JP2004523843A (en) | System, apparatus and method for diagnosing a flow system | |
| CN111651933B (en) | Industrial boiler fault early warning method and system based on statistical inference | |
| Hu et al. | A two-level intelligent alarm management framework for process safety | |
| De Simone et al. | LSTM-based failure prediction for railway rolling stock equipment | |
| Jin et al. | A methodology for equipment condition simulation and maintenance threshold optimization oriented to the influence of multiple events | |
| KR20220089853A (en) | Method for Failure prediction and prognostics and health management of renewable energy generation facilities using machine learning technology | |
| Tamssaouet et al. | Uncertainty quantification in system-level prognostics: application to Tennessee Eastman process | |
| Mirzaei et al. | A novel approach to repair time prediction and availability assessment of the equipment in power generation systems using fuzzy logic and Monte Carlo simulation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20131030 Termination date: 20200607 |