CN109872004A - A kind of equipment fault prediction based on fuzzy Bayesian network and health evaluating method - Google Patents
A kind of equipment fault prediction based on fuzzy Bayesian network and health evaluating method Download PDFInfo
- Publication number
- CN109872004A CN109872004A CN201910175092.XA CN201910175092A CN109872004A CN 109872004 A CN109872004 A CN 109872004A CN 201910175092 A CN201910175092 A CN 201910175092A CN 109872004 A CN109872004 A CN 109872004A
- Authority
- CN
- China
- Prior art keywords
- unit group
- breakdown
- equipment
- bayesian network
- health status
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The equipment fault prediction and health evaluating method that the invention discloses a kind of based on fuzzy Bayesian network, it include: the major failure of extract equipment, influence by the interaction between stand-alone device failure each in continuous production process to system is quantified as breakdown loss degree, fuzzy Bayesian network is constructed, realizes equipment fault prediction and health evaluating.The method of the present invention can make full use of the fault message of equipment, the representational failure of discovery, keep equipment fault prediction result more accurate, and health evaluating can be made more to meet truth by breakdown loss degree, design is reasonable, easy to operate, is with a wide range of applications.
Description
Technical field
The invention belongs to intelligent plant equipment construction technical fields, are related to equipment fault prediction and health evaluating method, tool
Body is related to a kind of equipment fault prediction based on fuzzy Bayesian network and health evaluating method.
Background technique
Intelligent plant equipment fault analysis and prediction are the demand of lean production, and the key task of construction intelligent plant
One of.As the integrated level and complexity of intelligent plant are higher and higher, system failure probability of happening and disabler probability also by
It gradually increases, and failure will cause great harm, the serious system that will lead to entirely fails and paralyses once occurring.If
The early stage that failure occurs, i.e., in the case where it does not also cause any damage to system, detection in time is out of order and is implemented reliable
Maintenance policy debugging, so that it may be largely avoided the generation of damage of product, systemic breakdown and catastrophic failure.
The equipment fault analysis of intelligent plant and prediction mainly include equipment fault type definition, the acquisition of equipment fault data
With processing, fault signature is extracted, fault diagnosis, failure predication, the links such as health state evaluation.Failure is that equipment cannot be completed to advise
Function or performance degradation are determined to the state for being unsatisfactory for prescribed requirement；Fault diagnosis utilizes equipment condition monitoring data, by means of intelligence
Energy diagnosis algorithm, diagnoses the equipment to have broken down, lays the foundation for Fault Isolation and trouble hunting；Failure predication according to
According to monitoring gained historical data and fault model, comprehensive analysis is carried out to various data and information resources, by it is based on model,
Failure prediction methods based on data-driven, based on probability statistics, position that pre- measurement equipment future may break down and general
Rate provides support for maintenance decision；Health evaluating is to decide whether startup separator diagnosis and equipment dimension with the health status of system
Shield.
Intelligent plant acquires the production process data of magnanimity, is based on the failure predication of data-driven (data-driven)
Method is gradually paid attention to and obtains fast development.There are many failure prediction methods based on monitoring data, such as Box at present
Equal propositions are used to handle autoregression integral moving average model (the Autoregressive Integrated of time series forecasting
Moving Average Model, ARIMA), which is substantially time series to be regarded as the overall process of internal association to grind
Study carefully, excludes exceptional value in time series, calculate the autocorrelation of time series data and characterize the hair in time series future with this
Exhibition trend, to achieve the purpose that the future value to data carries out short-term forecast.Singular spectrum analysis (Singular Spectrum
Analysis, SSA) it is a kind of time domain and the nonparametric fault time sequence prediction method that frequency domain combines, comprising insertion, unusual
Value is decomposed and three processes of reconstruct, the core of this method be effective component in abstraction sequence realize modeling to time series and
Prediction, time series can be non-linear, non-stationary and comprising noises.Support vector regression (Support Vector
Machines Regression, SVR) it is a kind of machine learning method for time series forecasting, this method is directed to limited sample
This, thought be realize structural risk minimization, by a Non-linear Kernel function by multidimensional input be mapped to it is more high-dimensional
Regressing calculation is executed after feature space, and then obtains the Nonlinear Mapping relationship with output-index, is realized to the pre- of time series
It surveys.
Above-mentioned failure prediction method cannot find continuous life since the information for excavating acquisition from device fault information is less
The interaction in process between each stand-alone device failure is produced, influence of the associated failure to system, Bu Nengfa are not accounted for
Existing major failure, therefore failure predication accuracy rate is lower.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a kind of equipment fault based on fuzzy Bayesian network
Prediction and health evaluating method, the major failure of extract equipment, by the phase between stand-alone device failure each in continuous production process
Influence of the interaction to system is quantified as breakdown loss degree, and building fuzzy Bayesian network realizes that equipment fault prediction is commented with health
Estimate.The method of the present invention can make full use of the fault message of equipment, the representational failure of discovery, make equipment fault prediction result
It is more accurate, and health evaluating can be made more to meet truth by breakdown loss degree, design is reasonable, easy to operate, has extensive
Application value.
This method chooses three kinds of failures being affected to equipment using failure transfer matrix as Bayesian network training
Sample set；Stand-alone device is divided into line production group, and calculates the rate of breakdown and failure damage of all line production groups
Mistake degree；Line production group, assembly line health status grade are determined based on fuzzy algorithmic approach；Use rate of breakdown, breakdown loss degree
Data and line production group, assembly line state of health data training Bayesian network model；Bayes is verified using test data
Network model, and model is applied in actual production, realize equipment fault prediction and health evaluating.
Specifically, present invention provide the technical scheme that
A kind of equipment fault detection method based on fuzzy Bayesian network, including the following steps:
A. training sample set W of the major failure as Bayesian network of N number of equipment is chosen, wherein every kind of equipment is chosen
Three kinds of major failures, the specific steps are as follows:
A1. the failure transition probability matrix of N number of equipment is constructed, the specific steps are as follows:
A11. the failure transfer number table of N number of equipment is constructed, element is a kind of event whithin a period of time of every kind of equipment in table
Barrier is transferred to the number C of other types failureij(i, j=1,2 ..., n), i are error code i, and j is error code j, and n indicates certain single machine
There is n kind error code in search time section in equipment；
Failure transfer is after a kind of failure occurs and repairs and to break down, twice failure possibility phase
It together may also be different；
A12. the failure transition probability matrix of N number of equipment is calculated, matrix element is every kind of equipment whithin a period of time by one
Kind failure is transferred to the probability P of other types failureij(i, j=1,2 ..., n),I is error code i, and j is event
Hinder code j, n indicates that n kind error code, such as P occurs in search time section in certain stand-alone device23Indicate that certain equipment is currently error code
2, next time is transferred to the probability of error code 3；
Wherein, N indicates the number of stand-alone device on detected assembly line Line；
A2. the error code specific gravity matrix G of all devices is constructedi(i=1,2 ..., N), i are i-th of equipment, specific steps
It is as follows:
A21. the number for calculating every kind of error code appearance of i-th of equipment accounts for the ratio of all error code frequency of occurrence of the equipment
Example constitutes the error code probability matrix R of i-th of equipmenti(i=1,2 ..., N), i are i-th of equipment；
A22. probability { the P on i-th of equipment fault transition probability matrix diagonal line is taken11,,P22,…,Pnn, constitute i-th
The error code persistence probability matrix W of a equipmenti(i=1,2 ..., N), i are i-th of equipment；
A23. error code probability matrix R is determined according to expertiseiWeight γ1With error code persistence probability matrix Wi's
Weight γ2；
A24. the error code specific gravity matrix G of i-th of equipment is calculatedi(i=1,2 ..., N), Gi=γ1×Ri+γ2×Wi, i
For i-th of equipment, γ1For error code probability matrix RiWeight, γ2For error code persistence probability matrix WiWeight；
A3. error code specific gravity matrix G is choseniMajor failure of the middle maximum three kinds of error code of specific gravity as i-th of equipment
Code；
A4. the training sample set W of Bayesian network is made of the major failure code of N number of equipment (i.e. 3*N error code)；
B. N number of stand-alone device on assembly line Line is divided into M line production group (abbreviation unit group), and calculates institute
There are the rate of breakdown and breakdown loss degree of unit group, the specific steps are as follows:
B1. according to the process of assembly line Line and equipment connection, N number of equipment is divided into an equipment or multiple phases
M combination of attached device, is denoted as M unit group；
B2. the rate of breakdown of M unit group is calculated, the specific steps are as follows:
B21. the number C that the major failure code of all devices in i-th of unit group occurs is determinediAnd assembly line Line
The total degree N that the major failure code of upper all devices occursn；
B22. the rate of breakdown P of i-th of unit group is calculatedi,I is i-th of unit group,
M is unit group number, CiFor the number that the major failure code of all devices in i-th of unit group occurs, NnFor on assembly line Line
The total degree that the major failure code of all devices occurs；
B3. the breakdown loss degree of M unit group is calculated, breakdown loss degree is yield brought by i-th of unit group failure
Loss ratio, the specific steps are as follows:
B31. the time t that each failure of i-th of unit group occurs is determinedjThe number k occurred with failure calculates i-th of list
The mean down time Δ t of tuplei,I is i-th of unit group, and M is unit group number,
tjFor the time that each failure occurs, k is the number that failure occurs；
B32. the part accumulation L of M unit group is calculatedi(i=1,2 ..., M) and part free quantity Hi(i=1,2 ...,
M), i is i-th of unit group, and M is unit group number, the specific steps are as follows:
When i-th of unit group of B321 breaks down, zero be deposited between (i-1)-th unit group and i-th of unit group is determined
Number of components h1And amount of parts h when breaking down between i-th cell group and i+1 unit group2；
B322 determines the speed Ψ t of i-th of unit group manufactured partsi(i=1,2 ..., M), i are i-th of unit group, and M is
Unit group number；
B323 calculates the part accumulation L of i-th of unit groupi:I is i-th of unit group, Δ ti
For the mean down time of i-th of unit group, Ψ tjFor the speed of j-th of unit group manufactured parts, h1It is sent out for i-th of unit group
The amount of parts being deposited in when raw failure between (i-1)-th unit group and i-th of unit group；
The part accumulation is i-th of unit group when breaking down, which does not work caused production loss；
B324 calculates the part free quantity H of i-th of unit groupi:I is i-th of unit
Group, n indicate that n kind error code, Δ t occurs in search time section in certain stand-alone deviceiFor i-th of unit group mean failure rate when
Between, Ψ tjFor the speed of j-th of unit group manufactured parts, h2I-th cell group and i+1 when breaking down for i-th of unit group
Amount of parts between unit group；
When the part free quantity is that i-th of unit group breaks down, the damage of yield caused by i+1 unit group does not work
It loses；
B33. the normal volume Q of M unit group when not breaking down is calculatedi,i
For i-th of unit group, M is unit group number, and n indicates that n kind error code, Δ t occurs in search time section in certain stand-alone devicei
For the mean down time of i-th of unit group, Ψ tjFor the speed of j-th of unit group manufactured parts；
B34. the coefficient lambda in breakdown loss degree formula is solved by regression analysis1、λ2, the specific steps are as follows:
It is breakdown loss degree δ that B341, which establishes dependent variable,i, independent variable is part accumulation/normal amount (m1) and the part free time
Amount/normal amount (m2) regression model: δi=λ1m1+λ2m2+ ξ, λ1For coefficient in breakdown loss degree formula, λ2For breakdown loss degree formula
Middle coefficient, m1For part accumulation/normal amount, m2For part free quantity/normal amount, ξ is error term；
B342 is idle by the breakdown loss degree, part accumulation, part of certain unit group in actually measured assembly line Line
Amount data substitute into the regression model established in step B341 and carry out linear fit, determine the coefficient lambda of regression model1、λ2Value,
Make the degree of fitting highest of model, and by gained λ1、λ2Value be applied to all unit groups breakdown loss degree calculate in；
B35. the breakdown loss degree δ of i-th of unit group is calculatedi,Wherein, LiFor i-th of unit
The part accumulation of group, HiFor the part free quantity of i-th of unit group, QiNormal production when not breaking down for i-th of unit group
Amount, λ1、λ2For coefficient determined by B3.4；
C. fuzzy algorithmic approach determination unit group, assembly line Line health status grade are based on, the specific steps are as follows:
C1. the rate of breakdown of unit group and breakdown loss degree are mapped as rate of breakdown grade and breakdown loss degree etc.
Grade, the specific steps are as follows:
C11. the membership function of fuzzy algorithmic approach is determined using assigning method；
C12. rate of breakdown is divided by small (S according to the value of gained rate of breakdown in B2 stepi) [0-0.2), in (Mi)
[0.2-0.4), big (Li) [0.4-1] three grades；
C13. breakdown loss degree is divided by small (s according to the value of gained breakdown loss degree in B3 stepi) [0-0.22), in
(mi) [0.22-0.24), big (li) [0.24-1] three grades；
C2. according to the rate of breakdown and breakdown loss degree grade of i-th of unit group, by the health status of the unit group point
For four grades: excellent (Si∨si), good (Si∨mi、Si∨li、Mi∨si), in (Mi∨mi、Mi∨li、Li∨si), poor (Li∨mi、
Li∨li)；
C3. by the grade of whole assembly line Line be divided into excellent (A), good (B), in (C), poor (D) four grades, partitioning standards
In the unit group for including for the number of the excellent state of whole assembly line Line unit group for including, even whole assembly line Line,
State is that the percentage that excellent unit group number accounts for the total unit group number of the assembly line is u, as u >=75%, assembly line Line
Health Category be excellent (A), when u>=50% and u<75%, the Health Category of assembly line Line is good (B), when u>=25% and
When u < 50%, the Health Category of assembly line Line be in (C), as u < 25%, the Health Category of assembly line Line is poor (D)；
D. using rate of breakdown, breakdown loss degree evidence and unit group, assembly line Line state of health data training shellfish
This network model of leaf obtains trained equipment fault detection Bayesian network model, the specific steps are as follows:
D1. training data is taken from the institute of the interior all devices of predicted assembly line Line longer period of time (>=365 days)
The raw data set that faulty data are formed, is divided into 70% training data and 30% test data for initial data, therefore
Barrier incidence, breakdown loss degree evidence and unit group, assembly line Line state of health data are passed through described by original training data
A, B, C method obtain；
D2. Bayesian network directed acyclic graph is constructed, the specific steps are as follows:
D21. topological diagram Bayesian network early period is constructed: by initial data-influence factor-health status hierarchical structure
And expertise determines Bayesian network topological diagram early period, variable relation are as follows: major failure code-> equipment, equipment-> failure
Incidence, equipment-> breakdown loss degree, rate of breakdown-> health status, breakdown loss degree-> health status；
D22. Bayesian network later period topological diagram is constructed: by assembly line health status-unit group state-rate of breakdown
Bayesian network later period topological diagram, variable relation are as follows: assembly line health status-> mono- are constructed with the hierarchical structure of breakdown loss degree
Tuple health status, unit group health status-> unit group rate of breakdown, unit group health status-> unit group failure damage
Mistake degree；
"-> " indicates the causality between two connected variables, the variable before "-> " be reason (also referred to as "-
> " father node of variable afterwards), the variable after "-> " is result (the also referred to as child node of variable before "-> ")；
D3. divide two layers and seek Bayesian network conditional probability table, the specific steps are as follows:
D31. assembly line Line health status is utilized, the conditional probability of M unit group health status is solved, by step C institute
Assembly line Line health status and unit group health status level data, calculate separately unit group health status be A,
B, the probability of C, D grade, unit group health status while the health status for calculating separately assembly line Line is A, B, C, D grade
For the probability of A, B, C, D grade, sought by condition probability formula in the health status of assembly line Line being A, B, C, D grade respectively
Under conditions of M unit group health status be A, B, C, D grade conditional probability；
D32. the health status for utilizing M unit group, solves the condition of M unit group rate of breakdown and breakdown loss degree
Probability, the health status level data and rate of breakdown and breakdown loss degree level data of the unit group obtained by step C, respectively
The probability that rate of breakdown is s, m, l grade is calculated, the probability that breakdown loss degree is S, M, L grade is calculated separately, calculates separately M
Rate of breakdown is the probability of s, m, l grade while a unit group health status is A, B, C, D grade, calculates separately M list
Breakdown loss degree is the probability of S, M, L grade while tuple health status is A, B, C, D grade, is distinguished by condition probability formula
The conditional probability that rate of breakdown under conditions of M unit group health status is A, B, C, D grade is s, m, l grade is sought,
Seeking breakdown loss degree under conditions of M unit group health status is A, B, C, D grade respectively is that the condition of S, M, L grade is general
Rate；
E. test data is applied to the Bayesian network model after training, to the M unit group on the same day on date D and assembly
Line Line health status makes assessment, and model is applied in actual production, the specific steps are as follows:
E1. using the historical data of a period of time (>=30 days) before date D, pass through A, B, step C the method obtains
The rate of breakdown and breakdown loss degree level data of M unit group；
E2. the fuzzy Bayes that the rate of breakdown of M unit group and the input of breakdown loss degree level data is constructed
Network model, the health status of model output unit group；
E3., unit group state of health data obtained by E2 is input to constructed fuzzy Bayesian network model, model is defeated
The health status of assembly line Line out；
E4. predict that resulting assembly line Line health status information judges the equipment fault situation of this day according to E3.
Compared with prior art, the beneficial effects of the present invention are:
Technological merit of the invention is the relationship considered in continuous production process between each stand-alone device failure, can and
When understanding producing line and line production group health status, failure is effectively predicted, convenient for targetedly taking dimension
It repairs measure and solves system problem in time, efficiently debug in time to maintenance personal, prevent harmfulness accident from occurring, improve dimension
Supportability is repaired to be of great significance.
Detailed description of the invention
Fig. 1 is the flow chart element of the equipment fault based on fuzzy Bayesian network and health evaluating method provided by the invention
Figure.
Fig. 2 is assembly line X in the embodiment of the present inventionaBayesian network topological diagram early period.
Fig. 3 is assembly line X in the embodiment of the present inventionaBayesian network later period topological diagram.
Specific embodiment
With reference to the accompanying drawing, the present invention, the model of but do not limit the invention in any way are further described by embodiment
It encloses.
Following embodiment is using certain household appliances enterprise in September, 2017 in July, 2018 assembly line XaData are (on the assembly line altogether
Have 9 stand-alone devices), the implementation process of prediction technique provided by the invention is described in detail.
Method flow diagram is as shown in Figure 1.The method of the present invention include: 1) using failure transfer matrix choose on equipment influence compared with
Three kinds of big failures are as Bayesian network training sample set；2) stand-alone device is divided into line production group, and calculates institute
There are the rate of breakdown and breakdown loss degree of line production group；3) line production group, assembly line health are determined based on fuzzy algorithmic approach
State grade；4) using rate of breakdown, breakdown loss degree evidence and line production group, assembly line state of health data training shellfish
This network model of leaf；5) Bayesian network model is verified using test data, and model is applied in actual production, realization is set
Standby failure predication and health evaluating.Specific implementation step is as follows:
1. choosing training sample set W of the 27 kinds of major failure codes of 9 equipment as Bayesian network, specific steps are such as
Under:
1.1. the failure transition probability matrix of 9 equipment is determined；
1.2. the error code specific gravity matrix G of all devices is determinedi(i=1,2 ..., 9), the specific steps are as follows:
1.2.1. the number for calculating every kind of error code appearance of i-th of equipment accounts for all error code of i-th of equipment and goes out occurrence
Several ratios constitutes the error code probability matrix R of i-th of equipmenti(i=1,2 ..., 9)；
1.2.2. probability { the P on i-th of equipment fault transition probability matrix diagonal line is taken11,,P22,…,Pnn, constitute the
The error code persistence probability matrix W of i equipmenti(i=1,2 ..., 9)；
1.2.3. error code probability matrix R is determinediWeight γ1With error code persistence probability matrix WiWeight γ2；
1.2.4. the error code specific gravity matrix G of i-th of equipment is calculatedi(i=1,2 ..., 9)；
1.3. error code specific gravity matrix G is choseniMajor failure of the middle maximum three kinds of error code of specific gravity as i-th of equipment
Code, the results are shown in Table 1 for Cass collection；
The major failure code of nine kinds of equipment in 1 embodiment of table
1.4. the training sample set W of Bayesian network is made of the major failure code of 9 equipment (i.e. 27 error code)；
2. 9 stand-alone devices on assembly line are divided into 4 line production groups (abbreviation unit group), and calculate all lists
The rate of breakdown and breakdown loss degree of tuple, the specific steps are as follows:
2.1. 9 equipment are divided into 4 unit groups, specific such as table 2:
Unit group in 2 embodiment of table divides table
2.2. the rate of breakdown of 4 unit groups is calculated, the specific steps are as follows:
2.2.1. the number C that the major failure code of all devices in i-th of unit group occurs is determinediWith studied assembly line
The total degree N that the major failure code of upper all devices occursn；
2.2.2. the rate of breakdown P of i-th of unit group is calculatedi,
2.3. the breakdown loss degree of 4 unit groups is calculated, the specific steps are as follows:
2.3.1. the time t that each failure of i-th of unit group occurs is determinedjThe number k occurred with failure, calculates all lists
The mean down time Δ t of tuplei,
2.3.2. the part accumulation L of 4 unit groups is calculatedi(i=1,2 ..., 4) and part free quantity Hi(i=1,
2 ..., 4), the specific steps are as follows:
2.3.2.1 when i-th of unit group breaks down, determination is deposited between (i-1)-th unit group and i-th of unit group
Amount of parts h1And amount of parts h when breaking down between i-th cell group and i+1 unit group2；
2.3.2.2 the speed of i-th of unit group manufactured parts is determined；
2.3.2.3 the part accumulation and part free quantity of i-th of unit group are calculated:
2.3.3. the normal volume of 4 unit groups when not breaking down is calculated,
2.3.4. the coefficient lambda in breakdown loss degree formula is solved by regression analysis1、λ2, the specific steps are as follows:
2.3.4.1 establishing dependent variable is breakdown loss degree δi, independent variable is part accumulation/normal amount (m1) and part sky
Not busy amount/normal amount (m2) regression model: δi=λ1m1+λ2m2+ξ；
2.3.4.2 according to actually measured assembly line XaIn the breakdown loss degree of certain unit group, part accumulation, part it is empty
Spare time amount data, determine parameter lambda1、λ2, make the degree of fitting highest of model, and by gained λ1、λ2Failure applied to all unit groups
During degree of loss calculates；
2.3.5. the breakdown loss degree δ of i-th of unit group is calculatedi,
3. being based on fuzzy algorithmic approach determination unit group, assembly line health status grade, the specific steps are as follows:
3.1. the rate of breakdown of unit group and breakdown loss degree are mapped as rate of breakdown grade and breakdown loss degree
Grade, the specific steps are as follows:
3.1.1. the membership function of fuzzy algorithmic approach is determined using assigning method；
3.1.2. rate of breakdown is divided by small (S according to the value of rate of breakdowni) [0-0.2), in (Mi)[0.2-0.4)、
(L greatlyi) [0.4-1] three grades, breakdown loss degree is divided by small (s according to the value of breakdown loss degreei) [0-0.22), in (mi)
[0.22-0.24), big (li) [0.24-1] three grades；
3.2. according to the rate of breakdown and breakdown loss degree grade of i-th of unit group, by the health status of the unit group
It is divided into four grades: excellent (Si∨si), good (Si∨mi、Si∨li、Mi∨si), in (Mi∨mi、Mi∨li、Li∨si), poor (Li∨
mi、Li∨li), it is specific such as table 3: the health status classification of line generation unit group in 3 embodiment of table
The grade of whole assembly line is divided into excellent by the excellent state number for the unit group for 3.3. including according to whole assembly line
(A), good (B), in (C), poor (D) four grades, it is specific such as table 4:
4 embodiment assembly line X of tableaHealth status table of grading
4. using rate of breakdown, breakdown loss degree evidence and unit group, assembly line XaState of health data trains pattra leaves
This network model, the specific steps are as follows: 4.1. constructs Bayesian network directed acyclic graph, the specific steps are as follows:
4.1.1. topological diagram Bayesian network early period is constructed: by initial data-influence factor-health status level knot
Structure and expertise determine Bayesian network topological diagram early period, variable relation are as follows: major failure code-> equipment, equipment-> event
Hinder incidence, equipment-> breakdown loss degree, rate of breakdown-> health status, breakdown loss degree-> health status, example structure
It builds such as attached drawing 2, equipment 1-9 is 9 stand-alone devices in example in figure, and the number for being directed toward 3 digits of each equipment is the equipment
The error code of 3 kinds of major failures, the rate of breakdown of rate of breakdown representative unit group, the failure damage of degree of loss representative unit group
Mistake degree, the health status of health status representative unit group and assembly line；
4.1.2. Bayesian network later period topological diagram is constructed: by assembly line health status-unit group (unit) state-event
The hierarchical structure for hindering incidence and breakdown loss degree constructs Bayesian network later period topological diagram, variable relation are as follows: assembly line health
State-> unit group health status, unit group health status-> unit group rate of breakdown, unit group health status-> unit
Group breakdown loss degree, example building such as attached drawing 3；
4.2. divide two layers and seek conditional probability table, the specific steps are as follows:
4.2.1. the conditional probability for solving 4 unit group health status, in assembly line XaHealth status be respectively A, B,
C, the health status of the lower 4 unit groups of D grade is respectively the conditional probability of A, B, C, D grade, sample result such as table 5:
Training data assembly line X obtained in 5 embodiment of tableaHealth status grade probability
The conditional probability of 4 unit groups in 6 embodiment of table
4.2.2. the conditional probability for solving 4 unit group rate of breakdown and breakdown loss degree, in 4 unit group health shapes
State be A, B, C, D grade under rate of breakdown be respectively s, m, l grade conditional probability, 4 unit group health status be A,
B, under C, D grade breakdown loss degree be respectively S, M, L grade conditional probability, the actual calculation such as table 7~10:
7 Unit1 rate of breakdown of table, breakdown loss degree conditional probability
8 Unit2 rate of breakdown of table, breakdown loss degree conditional probability
9 Unit3 rate of breakdown of table, breakdown loss degree conditional probability
10 Unit4 rate of breakdown of table, breakdown loss degree conditional probability
5. with the Bayesian network model after test data verifying training, to 4 unit groups on the same day on the 14th July in 2018
With assembly line XaHealth status makes assessment, and model is applied in actual production, the specific steps are as follows:
5.1. the data for utilizing on July 13rd, 14 days 1 April in 2018 are obtained by A, B, step C the method
Rate of breakdown and breakdown loss degree level data, as a result such as table 11:
11 unit group rate of breakdown of table, breakdown loss degree grade
5.2. the fuzzy Bayesian network that the rate of breakdown of unit group and the input of breakdown loss degree level data is constructed
The health status of network model output unit group, example export result such as table 12:
12 unit group health status of table
| Unit1 | Unit2 | Unit3 | Unit4 |
| D | B | C | B |
5.3., unit group state of health data obtained by E2 is input to constructed fuzzy Bayesian network model, model is defeated
The health status in assembly line on July 14th, 2018 out, sample result such as table 13:
13 unit group health status of table and assembly line XaHealth status
5.4. this day equipment fault situation is judged according to the 5.3 resulting assembly line health status informations of prediction.
The method of the invention realizes failure predications and health evaluating based on fuzzy Bayesian network, pass through 5 steps
Can be realized to failure be effectively predicted and assessment to health status, convenient for targetedly maintenance measures being taken to solve in time
System problem has very high economic benefit.
It is finally noted that the purpose for publicizing and implementing example is to help to further understand the present invention, but this field
Technical staff be understood that without departing from the spirit and scope of the invention and the appended claims, it is various replacement and repair
It is all possible for changing.Therefore, the present invention should not be limited to embodiment disclosure of that, and the scope of protection of present invention is to weigh
Subject to the range that sharp claim defines.
Claims (8)
1. a kind of equipment fault detection method based on fuzzy Bayesian network, characterized in that include:
The major failure of extract equipment；Using failure transfer matrix, each equipment chooses three kinds of major failures as Bayesian network
Network training sample set；
N number of stand-alone device on assembly line is divided into M line production group or unit group；By list each in continuous production process
Influence quantum chemical method of the interaction to system between machine equipment failure, rate of breakdown and breakdown loss as unit group
Degree；
The rate of breakdown of unit group and breakdown loss degree are mapped as rate of breakdown grade and breakdown loss degree grade；It is based on
The health status grade of fuzzy algorithmic approach determination unit group and the health status grade of assembly line；
Construct fuzzy Bayesian network；Use rate of breakdown, breakdown loss degree evidence and line production group, assembly line health shape
State data train Bayesian network model, obtain trained equipment fault detection Bayesian network model；
Bayesian network model is detected using trained equipment fault, assesses the unit group of certain time and the healthy shape of assembly line
State；
It is achieved in equipment fault detection.
2. the equipment fault detection method based on fuzzy Bayesian network as described in claim 1, characterized in that specifically choose N
Three kinds of failures of a equipment include the following steps: as Bayesian network training sample set W
A1. the failure transition probability matrix of N number of equipment is constructed, N indicates the number of stand-alone device on detected assembly line；Specifically hold
The following operation of row:
A11. the failure transfer number table of N number of equipment is constructed, element is that every kind of equipment is sent out by error code i whithin a period of time in table
The number C that raw failure transfer is error code jij；I, j=1,2 ..., n；N indicates that certain stand-alone device occurs in search time section
N kind error code；
A12. the failure transition probability matrix of N number of equipment is calculated, matrix element is every kind of equipment whithin a period of time by a kind of event
Hinder the probability P for transfer of breaking downij, i, j=1,2 ..., n,
A2. the error code specific gravity matrix G of all devices is constructedi；I=1,2 ..., N；Specifically perform the following operations:
A21. the number for calculating every kind of error code appearance of i-th of equipment accounts for the ratio of all error code frequency of occurrence of the equipment,
Constitute the error code probability matrix R of i-th of equipmenti；I=1,2 ..., N；
A22. probability { the P on i-th of equipment fault transition probability matrix diagonal line is taken11, P22..., Pnn, it constitutes i-th and sets
Standby error code persistence probability matrix Wi；I=1,2 ..., N；
A23. error code probability matrix R is determinediWeight γ1With error code persistence probability matrix WiWeight γ2；
A24. it is calculate by the following formula to obtain the error code specific gravity matrix G of i-th of equipmenti；I=1,2 ..., N；
Gi=γ1×Ri+γ2×Wi；
A3. error code specific gravity matrix G is choseniThe middle maximum three kinds of error code of specific gravity, the major failure code as i-th of equipment；
The major failure code of A4.N equipment is 3 × N number of error code, constitutes the training sample set W of Bayesian network.
3. the equipment fault detection method based on fuzzy Bayesian network as described in claim 1, characterized in that the calculating institute
There are the rate of breakdown and breakdown loss degree of unit group；Include the following steps:
B1. according to the process of assembly line and equipment connection, N number of equipment is divided into an equipment or multiple associated devices
M combination, is denoted as M unit group；
B2. the rate of breakdown for calculating M unit group, specifically performs the following operations:
B21. the number C that the major failure code of all devices in i-th of unit group occurs is determinediAnd all devices on assembly line
Major failure code occur total degree Nn；
B22. it is calculate by the following formula to obtain the rate of breakdown P of i-th of unit groupi:
B3. the breakdown loss degree of M unit group is calculated, breakdown loss degree is production loss brought by i-th of unit group failure
Ratio；Specifically perform the following operations:
B31. the time t that each failure of i-th of unit group occurs is determinedjThe number k occurred with failure, calculates all unit groups
Mean down time Δ ti, i=1,2 ..., M；
B32. the part accumulation L of M unit group is calculatediWith part free quantity Hi；I=1,2 ..., M；
B33. the normal volume Q of M unit group when not breaking down is calculatedi,
B34. the coefficient lambda in breakdown loss degree formula is solved by regression analysis1、λ2；
B35. the breakdown loss degree δ of i-th of unit group is calculatedi,Wherein, LiFor i-th unit group
Part accumulation, HiFor the part free quantity of i-th of unit group, QiNormal volume when not breaking down for i-th of unit group.
4. the equipment fault detection method based on fuzzy Bayesian network as claimed in claim 3, characterized in that step B32 institute
Stating calculating, specific step is as follows:
B321. when i-th of unit group breaks down, zero be deposited between (i-1)-th unit group and i-th of unit group is determined
Number of packages amount h1And amount of parts h when breaking down between i-th cell group and i+1 unit group2；
B322. the speed Ψ t of i-th of unit group manufactured parts is determinedi, i=1,2 ..., M；
B323. the part accumulation L of i-th of unit group is calculatediWith part free quantity Hi:
The part accumulation is i-th of unit group when breaking down, which does not work caused production loss；Part is empty
Spare time amount is i-th of unit group when breaking down, and i+1 unit group does not work caused production loss.
5. the equipment fault detection method based on fuzzy Bayesian network as claimed in claim 3, characterized in that step B34's
Specific step is as follows for solution:
B341. establishing dependent variable is breakdown loss degree δi, independent variable is part accumulation/normal amount and part free quantity/normal amount
Regression model: δi=λ1m1+λ2m2+ξ；
Wherein, part accumulation/normal amount is denoted as m1；Part free quantity/normal amount is denoted as m2；
B342. according to the breakdown loss degree, part accumulation and part free quantity number of certain unit group in actually measured assembly line
According to determining coefficient lambda1、λ2Value, make the degree of fitting highest of regression model；
Gained λ1、λ2Value can be used to all unit groups breakdown loss degree calculate in.
6. the equipment fault detection method based on fuzzy Bayesian network as described in claim 1, characterized in that described to be based on mould
Algorithm determination unit group, assembly line health status grade are pasted, is specifically performed the following operations:
C1. the rate of breakdown of unit group and breakdown loss degree are mapped as rate of breakdown grade and breakdown loss degree grade,
Specific step is as follows:
C11. the membership function of fuzzy algorithmic approach is determined using assigning method；
C12. rate of breakdown is divided by three grades according to the value of rate of breakdown, comprising:
It is small, it is denoted as Si, range be [0-0.2)；
In, it is denoted as Mi, range be [0.2-0.4)；
Greatly, it is denoted as Li, range is [0.4-1]；
C13. breakdown loss degree is divided by three grades according to the value of breakdown loss degree, comprising:
It is small, it is denoted as si, range be [0-0.22)；
In, it is denoted as mi, range be [0.22-0.24)；
Greatly, it is denoted as li, range is [0.24-1]；
C2. according to the rate of breakdown and breakdown loss degree grade of i-th of unit group, the health status of the unit group is divided into four
A grade, is respectively as follows:
It is excellent: Si∨si；
It is good: Si∨mi、Si∨liOr Mi∨si；
In: Mi∨mi、Mi∨liOr Li∨si；
Difference: Li∨miOr Li∨li；
The grade of assembly line is divided into four grades by the number of the excellent state for the unit group for C3. including according to assembly line: excellent/A,
Good/B, in/C, difference/D.
7. the equipment fault detection method based on fuzzy Bayesian network as described in claim 1, characterized in that sent out using failure
Raw rate, breakdown loss degree evidence and unit group, assembly line state of health data training Bayesian network model, specific steps are such as
Under:
D1. training data be taken from predicted assembly line for a period of time in all devices the faulty raw data set of institute, including
Rate of breakdown, breakdown loss degree evidence and unit group, assembly line state of health data；Raw data set is divided into 70%
Training data and 30% test data；
D2. Bayesian network directed acyclic graph is constructed, the specific steps are as follows:
D21. topological diagram Bayesian network early period is constructed:
Using initial data-influence factor-health status hierarchical structure, Bayesian network early period is determined further according to expertise
Topological diagram；Variable relation are as follows: major failure code-> equipment, equipment-> rate of breakdown, equipment-> breakdown loss degree, failure
Incidence-> health status, breakdown loss degree-> health status；
D22. Bayesian network later period topological diagram is constructed:
Bayesian network is constructed by the hierarchical structure of assembly line health status-unit group state-rate of breakdown and breakdown loss degree
Network later period topological diagram, variable relation are as follows: assembly line health status-> unit group health status, unit group health status-> unit
Group rate of breakdown, unit group health status-> unit group breakdown loss degree；
"-> " indicates the causality between two connected variables；
D3. divide two layers and acquire Bayesian network conditional probability table, the specific steps are as follows:
D31. assembly line health status is utilized, the conditional probability of the M different grades of health status of unit group is solved；Assembling
Under the different grades of health status of line, the conditional probability of the M different grades of health status of unit group；
D32. the health status of M unit group is utilized, the condition for solving M unit group rate of breakdown and breakdown loss degree is general
Rate, comprising: the conditional probability of the different grades of rate of breakdown under the M different grades of health status of unit group and in M
The conditional probability of different grades of breakdown loss degree under a different grades of health status of unit group.
8. the equipment fault detection method based on fuzzy Bayesian network as described in claim 1, characterized in that using training
Equipment fault detect Bayesian network model, the health status of assessment unit group and assembly line, the specific steps are as follows:
E1. the historical data of date D for the previous period is utilized, comprising: rate of breakdown and breakdown loss degree of M unit group etc.
Grade data；
E2. by the rate of breakdown and the trained fuzzy Bayesian network of breakdown loss degree level data input of M unit group
Model, the health status of model output unit group；
E3. unit group state of health data obtained by step E2 is input to fuzzy Bayesian network model, model exports assembly line
Health status；
E4. according to the resulting assembly line health status information of step E3, the equipment fault situation on the same day can be obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910175092.XA CN109872004B (en) | 2019-03-08 | 2019-03-08 | Equipment fault prediction and health assessment method based on fuzzy Bayesian network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910175092.XA CN109872004B (en) | 2019-03-08 | 2019-03-08 | Equipment fault prediction and health assessment method based on fuzzy Bayesian network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109872004A true CN109872004A (en) | 2019-06-11 |
| CN109872004B CN109872004B (en) | 2021-06-04 |
Family
ID=66920077
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910175092.XA Active CN109872004B (en) | 2019-03-08 | 2019-03-08 | Equipment fault prediction and health assessment method based on fuzzy Bayesian network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109872004B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8332337B2 (en) * | 2008-10-17 | 2012-12-11 | Lockheed Martin Corporation | Condition-based monitoring system for machinery and associated methods |
| WO2014001799A1 (en) * | 2012-06-26 | 2014-01-03 | Bae Systems Plc | Vehicle diagnostic system and method |
| CN105426651A (en) * | 2015-06-10 | 2016-03-23 | 北京交通大学 | Hidden trouble identification method based on composite fault chain inference |
| CN106250631A (en) * | 2016-08-03 | 2016-12-21 | 中国科学院西安光学精密机械研究所 | A kind of method for diagnosing faults based on fault test correlation matrix |
| CN107905990A (en) * | 2017-06-13 | 2018-04-13 | 武汉科技大学 | A kind of Fault Diagnosis of Hydraulic Pump system based on FUZZY H NETS and bayes method |
| CN108762242A (en) * | 2018-06-11 | 2018-11-06 | 宁波大学 | A kind of distributed fault detection method based on polylith canonical correlation analysis model |
-
2019
- 2019-03-08 CN CN201910175092.XA patent/CN109872004B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8332337B2 (en) * | 2008-10-17 | 2012-12-11 | Lockheed Martin Corporation | Condition-based monitoring system for machinery and associated methods |
| WO2014001799A1 (en) * | 2012-06-26 | 2014-01-03 | Bae Systems Plc | Vehicle diagnostic system and method |
| CN105426651A (en) * | 2015-06-10 | 2016-03-23 | 北京交通大学 | Hidden trouble identification method based on composite fault chain inference |
| CN106250631A (en) * | 2016-08-03 | 2016-12-21 | 中国科学院西安光学精密机械研究所 | A kind of method for diagnosing faults based on fault test correlation matrix |
| CN107905990A (en) * | 2017-06-13 | 2018-04-13 | 武汉科技大学 | A kind of Fault Diagnosis of Hydraulic Pump system based on FUZZY H NETS and bayes method |
| CN108762242A (en) * | 2018-06-11 | 2018-11-06 | 宁波大学 | A kind of distributed fault detection method based on polylith canonical correlation analysis model |
Non-Patent Citations (5)
| Title |
|---|
| DOS SANTOS, DAVI ANTONIO: "A Bayesian solution to the multiple composite hypothesis testing for fault diagnosis in dynamic systems", 《AUTOMATICA》 * |
| RUAN SUI: "dynamic multiple-fault diagnosis with imperfect tests", 《IEEE TRANSACTIONS ON SYSTEMS MAN AND CYBERNETICS PART A-SYSTEMS AND HUMANS》 * |
| 于重重: "决策树算法在农作物病虫害诊断中的应用", 《计算机工程与设计》 * |
| 何克磊: "飞行器健康评估和故障预测技术研究", 《中国优秀硕士学位论文全文数据库（工程科技Ⅱ辑）》 * |
| 姜珍: "SURROGATE PREPOSTERIOR ANALYSES FOR PREDICTING AND ENHANCING IDENTIFIABILITY IN MODEL", 《INTERNATIONAL JOURNAL FOR UNCERTAINTY QUANTIFICATION》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109872004B (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107358366B (en) | Distribution transformer fault risk monitoring method and system | |
| CN106444703B (en) | Dynamic equipment running status fuzzy evaluation and prediction technique based on fault mode probability of happening | |
| CN111237134B (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
| CN102208028B (en) | Fault predicting and diagnosing method suitable for dynamic complex system | |
| CN110222991B (en) | Metering device fault diagnosis method based on RF-GBDT | |
| CN102289590A (en) | Method for estimating operating state of SF6 high-voltage circuit breaker and intelligent system | |
| CN113156917A (en) | Power grid equipment fault diagnosis method and system based on artificial intelligence | |
| CN111598352A (en) | Concrete beam type bridge comprehensive evaluation method based on Bayesian network | |
| CN104504607A (en) | Method for diagnosing photovoltaic power station faults on the basis of fuzzy clustering algorithm | |
| Guo et al. | System operational reliability evaluation based on dynamic Bayesian network and XGBoost | |
| Li et al. | Preventive maintenance decision model of urban transportation system equipment based on multi-control units | |
| CN108898285A (en) | A kind of beer packaging production line operational efficiency quantitative evaluating method based on comentropy | |
| CN109523101A (en) | A kind of distribution Running State fuzzy synthetic appraisement method | |
| Kharlamova et al. | Evaluating machine-learning-based methods for modeling a digital twin of battery systems providing frequency regulation | |
| CN113361454B (en) | Non-supervision optimization-based deep learning non-invasive load monitoring method | |
| CN101930503A (en) | Method for computing importance in maintaining components of equipment | |
| Li et al. | Classification of the wind turbine components based on importance degrees: A three-way decision perspective | |
| CN109872004A (en) | A kind of equipment fault prediction based on fuzzy Bayesian network and health evaluating method | |
| Nie et al. | Research on AHP-based Multi-factor Medium Voltage Distribution Network Line Risk Quantitative Assessment Method | |
| Liu et al. | Real-time comprehensive health status assessment of hydropower units based on multi-source heterogeneous uncertainty information | |
| Zhang et al. | An evaluation method of health condition for wind turbine based on asymmetric proximity | |
| CN114661504A (en) | Operable and interpretable root cause positioning method for repeated occurrence type faults | |
| CN110336280B (en) | Power system cascading failure analysis method based on dictionary set acceleration | |
| Yu et al. | Short-term voltage stability assessment of multi-infeed HVDC systems based on JMIM and XGBoost | |
| Shan et al. | Root Cause Analysis of Failures for Power Communication Network Based on CNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |