CN110456199A - A kind of method for predicting residual useful life of multisensor syste - Google Patents
A kind of method for predicting residual useful life of multisensor syste Download PDFInfo
- Publication number
- CN110456199A CN110456199A CN201910750989.0A CN201910750989A CN110456199A CN 110456199 A CN110456199 A CN 110456199A CN 201910750989 A CN201910750989 A CN 201910750989A CN 110456199 A CN110456199 A CN 110456199A
- Authority
- CN
- China
- Prior art keywords
- useful life
- health
- particle
- predicting residual
- residual useful
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The present invention relates to electromechanical equipment health control technical fields, and disclose a kind of method for predicting residual useful life of multisensor syste, and specific implementation step is as follows: the improvement second order that S1 calculates sensing data arranges entropy；S2: the improved second order arrangement entropy of each sensing data is calculated according to calculation method described in S1 and is compared；S3: it selects improved second order arrangement entropy and the data of sensor is used for data fusion.The method for predicting residual useful life of the multisensor syste, it is proposed that a kind of improved second order arrangement entropy index measures the monotonicity trend of data, and multiple sensing datas are merged by data fusion method, construct the health status that the health indicator with good degradation trend and small threshold variance goes characterization equipment, again using the data of health indicator as the observation of tasteless particle filter, the health status of equipment is estimated and predicted, the final predicting residual useful life realized to more sensor-based systems.
Description
Technical field
The present invention relates to electromechanical equipment health control technical field, specially a kind of remaining life of multisensor syste is pre-
Survey method.
Background technique
With the rapid development of modern science and technology technology and the continuous improvement of functional requirement, the complexity of a large amount of electromechanical equipments,
Comprehensive and intelligent level is continuously improved, so that these electromechanical equipments become a large complicated more sensor-based systems, it is same with this
When, the reliability and safe running performance of these equipment also become more and more important, however, complicated operating condition and severe
Running environment, will lead to electromechanical equipment in the process of running has inevitable performance degradation, when the performance of electromechanical equipment is moved back
Change when being not enough to complete its function to equipment, will lead to equipment downtime even failure, and brings huge economic loss even people
Member's injures and deaths, as the key technology of prediction and health control, predicting residual useful life can be effectively predicted system and potentially degenerate
Gesture, estimate electromechanical equipment fault time, for ensure electromechanical equipment security service and reliability service be of great significance with it is practical
Value.
Currently, existing majority scholars study electromechanical equipment predicting residual useful life, and using distinct methods and
Technology predicts the remaining life of electromechanical equipment, some scholars are integrated with particle filter and realization the advantages of bond graph models
RUL prediction, and obtain higher precision and accurate fiducial limit；Some scholars establish improved autoregression model, and benefit
RUL is predicted with particle filter method, improves precision of prediction, still, the above method is all certain being only utilized in electromechanical equipment
One sensing data carries out predicting residual useful life, when the performance degradation mechanism of electromechanical equipment is apparent, and single sensor
When data can describe the degenerative process of electromechanical equipment, the above method is effectively, still, in practical projects, due to one
Often there are multiple sensors in electromechanical equipment, single sensing data is not enough to characterize the performance degradation of whole equipment, therefore, existing
There is method to be difficult to accurately predict the remaining life of multisensor electromechanical equipment.
Summary of the invention
The present invention provides a kind of method for predicting residual useful life of multisensor syste, electromechanics can accurately be predicted by, which having, is set
It the advantages of standby remaining life, solves the problems, such as to be previously mentioned in above-mentioned background technique.
The invention provides the following technical scheme: a kind of method for predicting residual useful life of multisensor syste, specific implementation step
It is rapid as follows:
The improvement second order that S1 calculates sensing data arranges entropy；
S2: the improved second order arrangement entropy of each sensing data is calculated according to calculation method described in S1 and is compared
Compared with；
S3: selecting improved second order arrangement entropy and the data of sensor are used for data fusion, and building can characterize more biographies
The health indicator of sensor system health status；
S4: it using health indicator as the observation of particle filter method, realizes and the health status of more sensor-based systems is predicted.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
Calculate sensing data improvement second order arrangement entropy the following steps are included:
S1.1: by i-th of sensing data xiIt maps to 2 dimension phase spaces to be reconstructed, obtains phase space as follows
Matrix:
Wherein A is the phase space matrix after reconstruct, and B is the submatrix in A, and n is the length of sensing data；
S1.2: a symbol sebolic addressing can be obtained for above-mentioned each state vector of phase space reconstruction matrix:
S (l)=[m1,m2], l=1,2 ..., n* (n-1)/2
Wherein l is state vector index, [m1,m2] it is the arrangement obtained by state vector numerical values recited, there are two types of may altogether
Property π1=[0,1] and π2=[1,0] takes [0,1] when two number numerical value are equal in state vector, compare each state to
Numerical values recited in amount obtains the Rankine-Hugoniot relations of each state vector；
S1.3: the corresponding arrangement π of all state vectors is calculated1,π2Probability；
S1.4: it calculates the improved second order of sensing data and arranges entropy.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
Calculate the corresponding arrangement π of all state vectors1,π2The calculation formula used of probability it is as follows:
Wherein # represents number.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
Calculate calculation formula used in the improved second order arrangement entropy of sensing data are as follows:
S2=abs (p (π1)log2[p(π1)]-p(π2)log2[p(π2)])
Wherein abs () is to take absolute value.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
Building can characterize multisensor syste health status health indicator specifically includes the following steps:
S3.1: the health indicator of t moment is constructed；
S3.2: solving biobjective scheduling problem by genetic algorithm, obtain optimal weight vectors ω *, calculates health and refers to
Target entirety monotonic trend and fault threshold variance.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
The building formula for constructing the health indicator of t moment is as follows:
Ht=x·,tω t=0,1,2 ..., ni
Wherein ω is weight vectors, x·,t∈R1×SIt is the measurement value matrix collected from S sensor；
The calculation formula of the whole monotonic trend for calculating health indicator are as follows:
s.t.ω′1S=1
Wherein S2ωIt (d) is that entropy is arranged by the improved second order of the weight vectors ω health index determined,It is by weight
The fault threshold variance for the health index that vector ω is determined, 1S∈RS×1It is the vector that value is all 1；
The fault threshold variance calculation formula is as follows:
Wherein Y ∈ RM×SIt is failure value matrix, M is the quantity of training unit.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
Particle filter method specifically includes the following steps:
S4.1: the state-space model of more sensor-based systems is constructed:
θk=f (θk-1,υk)
Hk=h (θk,νk)
Wherein θ is state vector, and f is nonlinear function, and υ is process variance, and H is health indicator, and h is non-Systems with Linear Observation
Function, ν are to measure variance；
S4.2: the importance density function of particle filter is generated with Unscented Kalman Filter；
S4.21: setting k=0 extracts N from prior distributionsA particle, and determine that weight is 1/Ns, then initialization is joined
Number:
Augmentation init state vector sum covariance matrix:
S4.22: Gauss point and weight calculation:
S4.23: the time updates:
S4.24: measuring value updates:
S4.3: it calculates sampling particle and updates particle weights:
S4.4: if number of effective particles is lower than given threshold value, resampling is carried out:
S4.5: state estimation:
Terminate to predict if k > T；
WhereinFor the particle of selection, the parameter of α and β for tasteless transformation, Wi (m)And Wi (c)Respectively first-order statistics characteristic
With the weight coefficient of second-order statistics, θk|k-1、Pk|k-1And Hk|k-1It is quantity of state, the one-step prediction of variance and measuring value, K respectivelyk
For filtering gain,WithRespectively the Unscented Kalman Filter last mean value arrived and variance；
S4.6: health status prediction is carried out to system using the prediction model that tasteless particle filter algorithm obtains；
S4.7: by the prediction to health status, the remaining life of more sensor-based systems is determined:
Wherein TH is the fault threshold of system, TRULIt is the remaining life of system, T is current time.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: when k≤
When T, k=k+1 is enabled, is then calculated by following steps:
S4.22: Gauss point and weight calculation:
S4.23: the time updates:
S4.24: measuring value updates:
S4.3: it calculates sampling particle and updates particle weights:
S4.4: if number of effective particles is lower than given threshold value, resampling is carried out:
S4.5: state estimation:
Terminate to predict if k > T
WhereinFor the particle of selection, the parameter of α and β for tasteless transformation, Wi (m)And Wi (c)Respectively first-order statistics characteristic
With the weight coefficient of second-order statistics, θk|k-1、Pk|k-1And Hk|k-1It is quantity of state, the one-step prediction of variance and measuring value, K respectivelyk
For filtering gain,WithRespectively the Unscented Kalman Filter last mean value arrived and variance；
S4.6: health status prediction is carried out to system using the prediction model that tasteless particle filter algorithm obtains；
S4.7: by the prediction to health status, the remaining life of more sensor-based systems is determined:
Wherein TH is the fault threshold of system, TRULIt is the remaining life of system, T is current time.
A kind of preferred embodiment of method for predicting residual useful life as multisensor syste of the present invention, in which: described
With the method that Unscented Kalman Filter generates the importance density function of particle filter first in sample phase, filtered with tasteless Kalman
Wave is that each particle calculates its mean value and covariance；Secondly it instructs to sample using the mean value and covariance.
The present invention have it is following the utility model has the advantages that
The method for predicting residual useful life of the multisensor syste proposes that a kind of improved second order arrangement entropy index measures data
Monotonicity trend, and multiple sensing datas are merged by data fusion method, construct with good degradation trend and compared with
The health indicator of small threshold variance goes the health status of characterization equipment, then using the data of health indicator as tasteless particle filter
Observation is estimated and is predicted to the health status of equipment, finally realizes the predicting residual useful life to more sensor-based systems.
Detailed description of the invention
Fig. 1 is the structural block diagram of the method for the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment 1
Referring to Fig. 1, a kind of method for predicting residual useful life of multisensor syste, the remaining longevity of the multisensor syste
Life prediction technique mainly includes using the health indicator of data fusion method characterization electromechanical equipment and using particle filter method reality
The prediction of existing electromechanical equipment health status, data fusion method are constructed to have and well be moved back by merging multiple sensing datas
The health indicator of change trend and small threshold variance goes the health status of characterization equipment, is referred to using data fusion method building health
It when mark, often allows health indicator that there is smaller improvement second order to arrange entropy, goes so that health indicator has good whole list
Tune trend, to improve the precision of prediction of electromechanical equipment remaining life, specific implementation step is as follows:
The improvement second order that S1 calculates sensing data arranges entropy, arranges entropy and data fusion method structure by improved second order
It builds health indicator and has good whole monotonic trend and lesser fault threshold error, so that predicting residual useful life precision is more
It is high；
S1.1: by i-th of sensing data xiIt maps to 2 dimension phase spaces to be reconstructed, obtains phase space as follows
Matrix:
Wherein A is the phase space matrix after reconstruct, and B is the submatrix in A, and n is the length of sensing data；
S1.2: a symbol sebolic addressing can be obtained for above-mentioned each state vector of phase space reconstruction matrix:
S (l)=[m1,m2], l=1,2 ..., n* (n-1)/2
Wherein l is state vector index, [m1,m2] it is the arrangement obtained by state vector numerical values recited, there are two types of may altogether
Property π1=[0,1] and π2=[1,0] takes [0,1] when two number numerical value are equal in state vector, compares each state vector
In numerical values recited, obtain the Rankine-Hugoniot relations of each state vector；
S1.3: the corresponding arrangement π of all state vectors is calculated1,π2Probability, it is as follows:
The wherein number that # is represented；
S1.4: it calculates the improved second order of sensing data and arranges entropy, improved second order arranges entropy calculation formula are as follows:
S2=abs (p (π1)log2[p(π1)]-p(π2)log2[p(π2)])
Wherein abs () is to take absolute value；
S2: the improved second order arrangement entropy of each sensing data is calculated according to calculation method described in S1 and is compared
Compared with；
S3: it selects improved second order arrangement entropy maximum and has the sensing data of good whole monotonic trend for counting
According to fusion, building can characterize the health indicator of multisensor syste health status,
S3.1: the health indicator of t moment is constructed, it is as follows that health indicator constructs formula:
Ht=x·,tω t=0,1,2 ..., ni
Wherein ω is weight vectors, x·,t∈R1×SIt is the measurement value matrix collected from S sensor；
S3.2: above-mentioned biobjective scheduling problem is solved by genetic algorithm, obtains optimal weight vectors ω*, it is ensured that it is strong
Kang Zhibiao has good whole monotonic trend and lesser fault threshold variance:
s.t.ω′1S=1
Wherein S2ωIt (d) is that entropy is arranged by the improved second order of the weight vectors ω health index determined,It is by weight
The fault threshold variance for the health index that vector ω is determined, 1S∈RS×1It is the vector that value is all 1, fault threshold variance calculates public
Formula is as follows:
Wherein Y ∈ RM×SIt is failure value matrix, M is the quantity of training unit；
S4: using health indicator as the observation of particle filter method, realizing and predict the health status of more sensor-based systems,
Particle filter is widely used and equipment residue because it can preferably solve non-linear, non-Gaussian filtering state estimation problem
Life prediction.
The present embodiment working principle are as follows: merge multiple sensing datas by using data fusion method, constructing has
The health indicator of good degradation trend and small threshold variance goes the health status of characterization equipment, later using health indicator as grain
The observation of sub- filtering method carries out health status prediction to system using the prediction model that particle filter algorithm obtains, finally
The remaining life of electromechanical equipment is calculated, data fusion method is used in conjunction with each other with particle filter method, further increases to machine
The precision of prediction of electric equipment remaining life.
Embodiment 2
Referring to Fig. 1, the difference of the present embodiment and above-described embodiment is: the used Unscented Kalman Filter of the present embodiment
The suggestion of particle filter is selected to be distributed, since particle filter algorithm is using prior distribution as suggesting Density Distribution letter
Number, when the registration very little of prior distribution and Posterior distrbutionp, the effect of particle filter algorithm is poor, causes with particle
Filtering algorithm predict electromechanical equipment remaining useful life when, precision of prediction can be reduced accordingly, thus the present embodiment use it is tasteless
Kalman filtering selects the suggestion of particle filter to be distributed, and is each particle with Unscented Kalman Filter first in sample phase
Its mean value and covariance are calculated, secondly instructs to sample using the mean value and covariance, is calculated using Unscented Kalman Filter equal
When value and covariance, newest observation information is utilized, so calculating electromechanical equipment residue to improve closer to Posterior distrbutionp
The accuracy when service life.
The method for predicting residual useful life of the multisensor syste mainly includes characterizing electromechanical equipment using data fusion method
Health indicator and the prediction of electromechanical equipment health status is realized using tasteless particle filter method, data fusion method is by melting
Multiple sensing datas are closed, the health indicator with good degradation trend and small threshold variance is constructed and goes the strong of characterization equipment
Health state when constructing health indicator using data fusion method, often allows health indicator to arrange with smaller improvement second order
Entropy is gone so that health indicator has good whole monotonic trend, so that the precision of prediction of electromechanical equipment remaining life is improved,
Specific implementation step is as follows:
The improvement second order that S1 calculates sensing data arranges entropy, arranges entropy and data fusion method structure by improved second order
It builds health indicator and has good whole monotonic trend and lesser fault threshold error, so that predicting residual useful life precision is more
It is high；
S1.1: by i-th of sensing data xiIt maps to 2 dimension phase spaces to be reconstructed, obtains phase space as follows
Matrix:
Wherein A is the phase space matrix after reconstruct, and B is the submatrix in A, and n is the length of sensing data；
S1.2: a symbol sebolic addressing can be obtained for above-mentioned each state vector of phase space reconstruction matrix:
S (l)=[m1,m2], l=1,2 ..., n* (n-1)/2
Wherein l is state vector index, [m1,m2] it is the arrangement obtained by state vector numerical values recited, there are two types of may altogether
Property π1=[0,1] and π2=[1,0] takes [0,1] when two number numerical value are equal in state vector, compare each state to
Numerical values recited in amount obtains the Rankine-Hugoniot relations of each state vector；
S1.3: the corresponding arrangement π of all state vectors is calculated1,π2Probability, it is as follows:
The wherein number that # is represented；
S1.4: it calculates the improved second order of sensing data and arranges entropy, improved second order arranges entropy calculation formula are as follows:
S2=abs (p (π1)log2[p(π1)]-p(π2)log2[p(π2)])
Wherein abs () is to take absolute value；
S2: the improved second order arrangement entropy of each sensing data is calculated according to calculation method described in S1 and is compared
Compared with；
S3: it selects improved second order arrangement entropy maximum and has the sensing data of good whole monotonic trend for counting
According to fusion, building can characterize the health indicator of multisensor syste health status,
S3.1: the health indicator of t moment is constructed, it is as follows that health indicator constructs formula:
Ht=x·,tω t=0,1,2 ..., ni
Wherein ω is weight vectors, x·,t∈R1×SIt is the measurement value matrix collected from S sensor；
S3.2: above-mentioned biobjective scheduling problem is solved by genetic algorithm, obtains optimal weight vectors ω*, it is ensured that it is strong
Kang Zhibiao has good whole monotonic trend and lesser fault threshold variance:
s.t.ω′1S=1
Wherein S2ωIt (d) is that entropy is arranged by the improved second order of the weight vectors ω health index determined,It is by weight
The fault threshold variance for the health index that vector ω is determined, 1S∈RS×1It is the vector that value is all 1, fault threshold variance calculates public
Formula is as follows:
Wherein Y ∈ RM×SIt is failure value matrix, M is the quantity of training unit；
S4: using health indicator as the observation of particle filter method, realizing and predict the health status of more sensor-based systems,
Particle filter method specifically includes following steps, and particle filter is because it can preferably solve non-linear, non-Gaussian filtering state
Estimation problem is widely used and equipment predicting residual useful life,
S4.1: the state-space model of more sensor-based systems is constructed:
θk=f (θk-1,υk)
Hk=h (θk,νk)
Wherein θ is state vector, and f is nonlinear function, and υ is process variance, and H is health indicator, and h is non-Systems with Linear Observation
Function, ν are to measure variance；
S4.2: the importance density function of particle filter is generated with Unscented Kalman Filter；
Particle filter algorithm is using prior distribution as density fonction is suggested, when prior distribution and Posterior distrbutionp
Registration very little when, the effect of particle filter can be bad, and therefore, the present invention selects grain using Unscented Kalman Filter
The suggestion distribution of son filtering is that each particle calculates its mean value and association side with Unscented Kalman Filter specifically in sample phase
Then difference instructs to sample using the mean value and covariance, because when calculating mean value and covariance with Unscented Kalman Filter, benefit
With newest observation information, so closer to Posterior distrbutionp；
S4.21: firstly, setting k=0, extracts N from prior distributionsA particle, and determine that weight is 1/Ns, then just
Beginningization parameter:
Augmentation init state vector sum covariance matrix:
S4.22: Gauss point and weight calculation:
S4.23: the time updates:
S4.24: measuring value updates:
S4.3: it calculates sampling particle and updates particle weights:
S4.4: if number of effective particles is lower than given threshold value, resampling is carried out:
S4.5: state estimation:
If meeting k≤T, k=k+1, return step S4.22 are enabled, otherwise, terminates prediction；
WhereinFor the particle of selection, the parameter of α and β for tasteless transformation, Wi (m)And Wi (c)Respectively first-order statistics characteristic
With the weight coefficient of second-order statistics, θk|k-1、Pk|k-1And Hk|k-1It is quantity of state, the one-step prediction of variance and measuring value, K respectivelyk
For filtering gain,WithRespectively the Unscented Kalman Filter last mean value arrived and variance；
S4.6: health status prediction is carried out to system using the prediction model that tasteless particle filter algorithm obtains；
S4.7: by the prediction to health status, the remaining life of more sensor-based systems is determined:
Wherein TH is the fault threshold of system, TRULIt is the remaining life of system, T is current time.
The working principle of the present embodiment are as follows: multiple sensing datas are merged using data fusion method first, pass through improvement
Second order arrangement entropy and data fusion method construct the health indicator with good degradation trend and small threshold variance and remove table
The health status for levying equipment, later using health indicator as the observation of tasteless particle filter method, using tasteless particle filter
Algorithm obtains prediction model and carries out health status prediction to system, and data fusion method makes with particle filter method mutual cooperation
With further increasing the precision of prediction to electromechanical equipment remaining life.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (9)
1. a kind of method for predicting residual useful life of multisensor syste, it is characterised in that: specific implementation step is as follows:
The improvement second order that S1 calculates sensing data arranges entropy；
S2: the improved second order arrangement entropy of each sensing data is calculated according to calculation method described in S1 and is compared；
S3: it selects improved second order arrangement entropy and the data of sensor is used for data fusion, building can characterize multisensor
The health indicator of system health status；
S4: it using health indicator as the observation of particle filter method, realizes and the health status of more sensor-based systems is predicted.
2. the method for predicting residual useful life of multisensor syste according to claim 1, it is characterised in that: the calculating passes
Sensor data improvement second order arrangement entropy the following steps are included:
S1.1: by i-th of sensing data xiIt maps to 2 dimension phase spaces to be reconstructed, obtains phase space matrix as follows:
Wherein A is the phase space matrix after reconstruct, and B is the submatrix in A, and n is the length of sensing data；
S1.2: a symbol sebolic addressing can be obtained for above-mentioned each state vector of phase space reconstruction matrix:
S (l)=[m1,m2], l=1,2 ..., n* (n-1)/2
Wherein l is state vector index, [m1,m2] it is the arrangement obtained by state vector numerical values recited, there are two types of possibility π altogether1
=[0,1] and π2=[1,0] takes [0,1] when two number numerical value are equal in state vector, compares in each state vector
Numerical values recited, obtain the Rankine-Hugoniot relations of each state vector；
S1.3: the corresponding arrangement π of all state vectors is calculated1,π2Probability；
S1.4: it calculates the improved second order of sensing data and arranges entropy.
3. the method for predicting residual useful life of multisensor syste according to claim 2, it is characterised in that: the calculating institute
The corresponding arrangement π of stateful vector1,π2The calculation formula used of probability it is as follows:
Wherein # represents number.
4. the method for predicting residual useful life of multisensor syste according to claim 3, it is characterised in that: the calculating passes
Calculation formula used in the improved second order arrangement entropy of sensor data are as follows:
S2=abs (p (π1)log2[p(π1)]-p(π2)log2[p(π2)])
Wherein abs () is to take absolute value.
5. the method for predicting residual useful life of multisensor syste according to claim 4, it is characterised in that: the building energy
Characterize multisensor syste health status health indicator specifically includes the following steps:
S3.1: the health indicator of t moment is constructed；
S3.2: biobjective scheduling problem is solved by genetic algorithm, optimal weight vectors ω * is obtained, calculates health indicator
Whole monotonic trend and fault threshold variance.
6. the method for predicting residual useful life of multisensor syste according to claim 5, it is characterised in that: the building t
The building formula of the health indicator at moment is as follows:
Ht=x·,tω t=0,1,2 ..., ni
Wherein ω is weight vectors, x·,t∈R1×SIt is the measurement value matrix collected from S sensor；
The calculation formula of the whole monotonic trend for calculating health indicator are as follows:
s.t.ω′1S=1
Wherein S2ωIt (d) is that entropy is arranged by the improved second order of the weight vectors ω health index determined,It is by weight vectors
The fault threshold variance for the health index that ω is determined, 1S∈RS×1It is the vector that value is all 1；
The fault threshold variance calculation formula is as follows:
Wherein Y ∈ RM×SIt is failure value matrix, M is the quantity of training unit.
7. the method for predicting residual useful life of -6 any multisensor systes according to claim 1, it is characterised in that: described
Particle filter method specifically includes the following steps:
S4.1: the state-space model of more sensor-based systems is constructed:
θk=f (θk-1,υk)
Hk=h (θk,νk)
Wherein θ is state vector, and f is nonlinear function, and υ is process variance, and H is health indicator, and h is non-Systems with Linear Observation function, ν
It is to measure variance；
S4.2: the importance density function of particle filter is generated with Unscented Kalman Filter；
S4.21: setting k=0 extracts N from prior distributionsA particle, and determine that weight is 1/Ns, then initiation parameter:
Augmentation init state vector sum covariance matrix:
S4.22: Gauss point and weight calculation:
S4.23: the time updates:
S4.24: measuring value updates:
S4.3: it calculates sampling particle and updates particle weights:
S4.4: if number of effective particles is lower than given threshold value, resampling is carried out:
S4.5: state estimation:
Terminate to predict if k > T；
WhereinFor the particle of selection, the parameter of α and β for tasteless transformation, Wi (m)And Wi (c)Respectively first-order statistics characteristic and two
The weight coefficient of rank statistical property, θk|k-1、Pk|k-1And Hk|k-1It is quantity of state, the one-step prediction of variance and measuring value, K respectivelykFor filter
Wave gain,WithRespectively the Unscented Kalman Filter last mean value arrived and variance；
S4.6: health status prediction is carried out to system using the prediction model that tasteless particle filter algorithm obtains；
S4.7: by the prediction to health status, the remaining life of more sensor-based systems is determined:
Wherein TH is the fault threshold of system, TRULIt is the remaining life of system, T is current time.
8. the method for predicting residual useful life of multisensor syste according to claim 6, it is characterised in that: as k≤T,
K=k+1 is enabled, is then calculated by following steps:
S4.22: Gauss point and weight calculation:
S4.23: the time updates:
S4.24: measuring value updates:
S4.3: it calculates sampling particle and updates particle weights:
S4.4: if number of effective particles is lower than given threshold value, resampling is carried out:
S4.5: state estimation:
Terminate to predict if k > T
WhereinFor the particle of selection, the parameter of α and β for tasteless transformation, Wi (m)And Wi (c)Respectively first-order statistics characteristic and two
The weight coefficient of rank statistical property, θk|k-1、Pk|k-1And Hk|k-1It is quantity of state, the one-step prediction of variance and measuring value, K respectivelykFor filter
Wave gain,WithRespectively the Unscented Kalman Filter last mean value arrived and variance；
S4.6: health status prediction is carried out to system using the prediction model that tasteless particle filter algorithm obtains；
S4.7: by the prediction to health status, the remaining life of more sensor-based systems is determined:
Wherein TH is the fault threshold of system, TRULIt is the remaining life of system, T is current time.
9. the method for predicting residual useful life of multisensor syste according to claim 8, it is characterised in that: described with tasteless
Kalman filtering generates the method for the importance density function of particle filter first in sample phase, is every with Unscented Kalman Filter
A particle calculates its mean value and covariance；Secondly it instructs to sample using the mean value and covariance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910750989.0A CN110456199B (en) | 2019-08-14 | 2019-08-14 | Residual life prediction method of multi-sensor system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910750989.0A CN110456199B (en) | 2019-08-14 | 2019-08-14 | Residual life prediction method of multi-sensor system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110456199A true CN110456199A (en) | 2019-11-15 |
| CN110456199B CN110456199B (en) | 2021-03-23 |
Family
ID=68486659
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910750989.0A Active CN110456199B (en) | 2019-08-14 | 2019-08-14 | Residual life prediction method of multi-sensor system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110456199B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111222290A (en) * | 2020-01-13 | 2020-06-02 | 浙江工业大学 | Large-scale equipment residual service life prediction method based on multi-parameter feature fusion |
| CN111553590A (en) * | 2020-04-27 | 2020-08-18 | 中国电子科技集团公司第十四研究所 | Radar embedded health management system |
| CN111751508A (en) * | 2020-05-12 | 2020-10-09 | 北京华科仪科技股份有限公司 | Performance evaluation prediction method and system for life cycle of water quality sensor |
| CN112613646A (en) * | 2020-12-08 | 2021-04-06 | 上海交通大学烟台信息技术研究院 | Equipment state prediction method and system based on multi-dimensional data fusion |
| CN112836381A (en) * | 2021-02-19 | 2021-05-25 | 震兑工业智能科技有限公司 | Multi-source information-based ship residual life prediction method and system |
| CN113627088A (en) * | 2021-08-23 | 2021-11-09 | 上海交通大学 | Machine performance degradation evaluation method and system based on gene programming and data fusion |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104792529A (en) * | 2015-04-12 | 2015-07-22 | 北京化工大学 | Rolling bearing life prediction method based on state-space model |
| CN106845866A (en) * | 2017-02-27 | 2017-06-13 | 四川大学 | Equipment method for predicting residual useful life based on improved particle filter algorithm |
| CN109376401A (en) * | 2018-09-29 | 2019-02-22 | 西安交通大学 | A kind of adaptive multi-source information preferably with the mechanical method for predicting residual useful life that merges |
| CN109948860A (en) * | 2019-03-26 | 2019-06-28 | 哈工大机器人（合肥）国际创新研究院 | A kind of mechanical system method for predicting residual useful life and system |
-
2019
- 2019-08-14 CN CN201910750989.0A patent/CN110456199B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104792529A (en) * | 2015-04-12 | 2015-07-22 | 北京化工大学 | Rolling bearing life prediction method based on state-space model |
| CN106845866A (en) * | 2017-02-27 | 2017-06-13 | 四川大学 | Equipment method for predicting residual useful life based on improved particle filter algorithm |
| CN109376401A (en) * | 2018-09-29 | 2019-02-22 | 西安交通大学 | A kind of adaptive multi-source information preferably with the mechanical method for predicting residual useful life that merges |
| CN109948860A (en) * | 2019-03-26 | 2019-06-28 | 哈工大机器人（合肥）国际创新研究院 | A kind of mechanical system method for predicting residual useful life and system |
Non-Patent Citations (5)
| Title |
|---|
| DAVID L.HALL,ET AL: "An Introduction to Multisensor Data Fusion", 《PROCEEDINGS OF THE IEEE》 * |
| KAIBO LIU,ET AL: "A Data-Level Fusion Model for Developing", 《IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING》 * |
| LINXIA LIAO,ET AL: "Enhanced Restricted Boltzmann Machine With", 《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》 * |
| 刘连胜: "基于信息熵测度的数据驱动剩余寿命在线预测方法研究", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 周俊: "数据驱动的航空发动机剩余使用寿命预测方法研究", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111222290A (en) * | 2020-01-13 | 2020-06-02 | 浙江工业大学 | Large-scale equipment residual service life prediction method based on multi-parameter feature fusion |
| CN111222290B (en) * | 2020-01-13 | 2024-04-09 | 浙江工业大学 | Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment |
| CN111553590A (en) * | 2020-04-27 | 2020-08-18 | 中国电子科技集团公司第十四研究所 | Radar embedded health management system |
| CN111751508A (en) * | 2020-05-12 | 2020-10-09 | 北京华科仪科技股份有限公司 | Performance evaluation prediction method and system for life cycle of water quality sensor |
| CN112613646A (en) * | 2020-12-08 | 2021-04-06 | 上海交通大学烟台信息技术研究院 | Equipment state prediction method and system based on multi-dimensional data fusion |
| CN112836381A (en) * | 2021-02-19 | 2021-05-25 | 震兑工业智能科技有限公司 | Multi-source information-based ship residual life prediction method and system |
| CN112836381B (en) * | 2021-02-19 | 2023-03-14 | 震兑工业智能科技有限公司 | Multi-source information-based ship residual life prediction method and system |
| CN113627088A (en) * | 2021-08-23 | 2021-11-09 | 上海交通大学 | Machine performance degradation evaluation method and system based on gene programming and data fusion |
| CN113627088B (en) * | 2021-08-23 | 2024-04-09 | 上海交通大学 | Machine performance degradation evaluation method and system based on gene programming and data fusion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110456199B (en) | 2021-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110456199A (en) | A kind of method for predicting residual useful life of multisensor syste | |
| Craven et al. | Ordinary least-squares regression | |
| CN108628281B (en) | Abnormality detection system and abnormality detection method | |
| CN102789545B (en) | Based on the Forecasting Methodology of the turbine engine residual life of degradation model coupling | |
| CN107451102B (en) | Method for predicting concentration of butane at bottom of debutanizer tower based on improved self-training algorithm semi-supervised Gaussian process regression soft measurement modeling | |
| Feng et al. | A state-space-based prognostic model for hidden and age-dependent nonlinear degradation process | |
| CN103344923B (en) | EKF (extended Kalmar filter)-method and NSDP-AR-model-based method for predicting cycling life of fusion-type lithium ion battery | |
| CN105300692B (en) | A kind of bearing failure diagnosis and Forecasting Methodology based on expanded Kalman filtration algorithm | |
| CN112347692B (en) | Method and device for realizing battery predictive maintenance of uninterruptible power supply and electronic device | |
| CN102879728A (en) | Health evaluation index and failure predication method for DC (Direct Current)-DC convertor | |
| EP3133406B1 (en) | Device and method for estimating operation states of individual electrical devices | |
| CN106909756A (en) | A kind of rolling bearing method for predicting residual useful life | |
| Schenk et al. | Bayesian estimation and prediction with multiply Type-II censored samples of sequential order statistics from one-and two-parameter exponential distributions | |
| CN111651708B (en) | Early warning threshold setting method for abnormal identification of dam safety monitoring data | |
| Wei et al. | Forecasting the geomagnetic activity of the Dst index using multiscale radial basis function networks | |
| Park et al. | Missing-insensitive short-term load forecasting leveraging autoencoder and LSTM | |
| Bartram et al. | Integration of heterogeneous information in SHM models | |
| CN113158471A (en) | Degradation equipment residual life prediction method considering measurement uncertainty | |
| Ma et al. | Reliability estimation from two types of accelerated testing data considering measurement error | |
| CN114290960A (en) | Method and device for acquiring battery health degree of power battery and vehicle | |
| Chen et al. | Remaining useful life prediction for complex systems with multiple indicators based on particle filter and parameter correlation | |
| CN114896861A (en) | Rolling bearing residual life prediction method based on square root volume Kalman filtering | |
| Crowder et al. | Small sample properties of an adaptive filter applied to low volume SPC | |
| CN108829983A (en) | Equipment method for predicting residual useful life based on more hidden state fractional Brownian motions | |
| CN108108839A (en) | Grid information system equipment state method for early warning based on reverse Fuzzy Level Analytic Approach |
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 |