CN104978459B - A kind of IGBT method for predicting residual useful life - Google Patents
A kind of IGBT method for predicting residual useful life Download PDFInfo
- Publication number
- CN104978459B CN104978459B CN201510396832.4A CN201510396832A CN104978459B CN 104978459 B CN104978459 B CN 104978459B CN 201510396832 A CN201510396832 A CN 201510396832A CN 104978459 B CN104978459 B CN 104978459B
- Authority
- CN
- China
- Prior art keywords
- hidden layer
- data
- igbt
- input data
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of IGBT method for predicting residual useful life, by phase space framing reconfiguration technique, on the basis of differential entropy rate phase space reconstruction, to the phase space that has reconstructed with the input signal vector arrangement of Volterra series；The correlation exported in view of input data with target, the optimal selection of each frame input data is carried out, preferably input data is used as the input of model to the present invention to select in input vector using the forward-backward algorithm algorithm of current comparative maturity and minimum angle regression algorithm；On the basis of original ELM models, add multiple response sparse regression algorithm and extraction method crops hidden layer node useless or that effect is seldom one by one, and using three kinds of neuron activation functions of mixing, so that the network established has more robustness and generalization；The present invention sufficiently considers that different inputs to the otherness of forecast model, devise the forecast model that a kind of adaptive algorithm dynamically updates each group of input data, greatly improve the precision of prediction.
Description
Technical field
The invention belongs to novel power semiconductor device reliability analysis technical field, more specifically, a kind of IGBT
Method for predicting residual useful life.
Background technology
Insulated gate bipolar transistor (IGBT, Insulated Gate Bipolar Transistor) is opened as power
Device is closed, it has many advantages such as current carrying density is big, saturation voltage drop is low, is widely used to generation of electricity by new energy, high pressure at present
In many key areas such as transmission of electricity.
As the core of system, IGBT reliability effect the operation stability of whole system equipment, and this causes
Research to IGBT residual life (RUL, Remaining Useful Life) Forecasting Methodology becomes very necessary, and it has
Some significance below：(1) be obtain IGBT reliability informations important channel, may further be realize system monitor on-line
Foundation is provided with health control；(2) contribute to promote manufacturer to carry out IGBT process modification (the new material of introducing and improvement envelope
Dress technology)；(3) be advantageous to preferably design accelerated aging test to obtain more accurate aging data；(4) can be achieved to regard feelings
Maintenance, terminal user is set to obtain the more life informations of IGBT to reduce the input to system maintenance.
The research of the existing method for predicting residual useful life to IGBT is broadly divided into two major classes:Based on the pre- of physical model driving
Survey technology and the Predicting Technique based on data-driven.Although the Predicting Technique based on physical model driving can be from the angle of material
IGBT residual life information is stated, but needs to have the intrinsic propesties of device manufacture material with industrial manufacturing process deep enough
Understand, but the physical model obtained often lacks enough precision, it is non-linear between the process image parameter easy to be lost of modeling
Relation, error is caused to increase.On the other hand, constructed physical model is often closely related with device concrete model, this and city
The product category that IGBT is skyrocketed through on field produces contradiction, thus such method necessarily brings the hysteresis quality in Forecasting Methodology.And
Predicting Technique based on data-driven is the mapping relations inputted from IGBT historical aging data learning between output, then
Non-linear, the nontransparent and non-model for special object is internally established, to the remaining lifetime value of calculating device.If energy
Accurate IGBT forecast models are enough established, precision of prediction will be greatly enhanced, so as to further reduce prediction error.But current base
It is also fewer in the IGBT method for predicting residual useful life of data-driven, and the degree of accuracy predicted need to be improved.
The content of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of IGBT method for predicting residual useful life, to carry
The precision of prediction of high residual life.
For achieving the above object, IGBT method for predicting residual useful life of the present invention, it is characterised in that including following step
Suddenly：
(1), by accelerated life test, IGBT collection emitter-base bandgap grading saturation voltage drop aging data time serieses is obtained, obtain data
Collect DS={ x (1), x (2) ..., x (D) }, wherein D is obtained data amount check；
(2), first, for data set DS, optimal Embedded dimensions d and time delay are determined according to the method for differential entropy rate
τ；Then, framing reconstruct is carried out to data set DS data sets using the function windowize in Matlab, and each frame data is divided
D dimensional feature spaces are not mapped to, so as to obtain n (n=D-d τ) individual data frame { (Xt,Yt), wherein t=1,2 ..., n, using as
Input data and the target output of training pattern, wherein：
Xt=1, x (t), x (t+ τ) ..., x (t+ (d-1) τ),
x2(t), x (t) x (t+ τ) ... x (t) x (t+ (d-1) τ),
x2(t+ τ), x (t+ τ) x (t+2 τ) ..., x (t+ τ) x (t+ (d-1) τ),
x2(t+2 τ), x (t+2 τ) x (t+3 τ) ..., x (t+2 τ) x (t+ (d-1) τ)
...,
x2(t+ (d-2) τ), x (t+ (d-2) x (t+ (d-1) τ)
x2(t+(d-1)τ)}；
Yt=x (t+d τ)
(3) Forward-backward algorithm (FB, Forward-Backward algorithm) or minimum angle regression algorithm, are utilized
Every frame input data X that (LARS, Least Angle Regression algorithm) obtains to step (2)tMake optimal choosing
Select, select and export Y with corresponding targettM higher input data of correlation, note selection after every frame input data for：
Xt'={ 1, x (t+c1τ),x(t+c2τ),…,x(t+cbτ),…,x(t+ceτ)x(t+cfτ)}；cb,ce,cf∈{0,
1 ..., d-1 }, and ce≤cf；
Then per frame input data Xt' dimension size (element number) be m, can be expressed as again：
Xt'=[xt,1,xt,2,…,xt,m]T(t=1,2 ..., n)；
(4) extreme learning machine (ELM) model for containing N (N ＜ D) individual hidden layer node, is initialized；The number of input layer
According to the input data X obtained for step (3)t' (t=1,2 ..., n)；
In section [- 1,1] interior random initializtion ELM input weight coefficient matrixWith threshold coefficient matrixWherein s=1,2 ..., N；I=1,2 ..., m；
According to the input weight coefficient matrix of initializationWith threshold coefficient matrixCalculate hidden layer output matrixWherein network hidden layer output vectorFor：
G represents excitation function；
(5), the vector obtained according to step (4)And the corresponding target output Y obtained in step (2)t, utilize polyphony
Sparse regression algorithm (MRSR, Multiresponse Sparse Regression) is answered to vectorElement be hidden layer section
Point is rearranged, and the hidden layer node after note sequence is：
Wherein, 1≤j of subscripti≤ N is the hidden layer node sequence number before sequence, on
It is the hidden layer node precedence sequence number after sequence to mark 1≤i≤N；According to the order sequenced, l hidden layer node before extraction, wherein
L ＜ N；
(6) weight coefficient and threshold coefficient, input weights system, are inputted according to corresponding to step (5) extracts hidden layer node
Number and threshold coefficient matrix are updated to w=(w respectivelys,i)l×mWith θ=θsS=1,2 ..., l；I=1,2 ..., m, and to taking out
After taking (beta pruning), hidden layer output matrix H=[h are recalculated1,h2,…,hj,…,hn], wherein vector
(7) the network output weights after beta pruning, are solved and prediction result, its process are as follows：
7.1) the D+1 data, is predicted, i.e. during prediction data x (D+1), according to forecast model knowable to step (2) and (3)
Original input data frame be Xt, now, t=D+1-d τ；After selection is inputted, input data frame is X after note selectiont', t=D+
1-d τ, then the network hidden layer output vector at the moment can be calculated i.e. t=D+1-d τ moment
7.2), calculating network hidden layer output vector htThe hidden layer output matrix H=[h obtained with step (6)1,
h2,…,hj,…,hn] in each vectorial hj, j=1,2 ..., n Euclidean distance S；
7.3), to network hidden layer output vector h in Euclidean distance StWith hidden layer output matrix H each vectorial hj,j
=1,2 ..., n distance value are ranked up, and are found out from hidden layer output matrix H closest to network hidden layer output vector ht
Z, z >=l vector form new matrixAnd corresponding target outputFrom
And calculate corresponding output weights
T=D+1-d τ moment, the predicted value of forecast model are obtained by the output weights：
(8), the predicted value that judgment step (7) obtainsWhether reach IGBT performance failure threshold value (APT,
Acceptable Performance Threshold), if reached, the predicted value position is that D+1 records as critical point,
Stop prediction, if be not reaching to, by predicted valueIGBT collection emitter-base bandgap grading saturation voltage drop aging data time serieses are added to, are obtained
To data set DS={ x (2), x (3) ..., x (D+1) }, then predicted value is obtained according to the method for step (1)~(7)And sentence
The disconnected performance failure threshold value for whether reaching IGBT, if it is not, continuously adding IGBT collection emitter-base bandgap grading saturation voltage drop aging datas
Time series, data set DS={ x (3), x (4) ..., x (D+2) } is obtained, so repeat, predicted untill reaching, and by this
It is worth position D+W to record as critical point, wherein, W is predicted valueReach the prediction number of IGBT performance failure threshold value.
The object of the present invention is achieved like this.
IGBT method for predicting residual useful life of the present invention, by phase space framing reconfiguration technique, reconstructed in differential entropy rate mutually empty
Between on the basis of, in order to pursue higher precision of prediction, change and consecutive number be predicted as to IGBT collection emitter-base bandgap grading saturation voltage drop data
According to the prediction of difference, to the phase space that has reconstructed with the input signal vector arrangement of Volterra series；In view of input data
With the correlation of target output, i.e., some inputs do not work to prediction or play very little, carry out each frame input data
Optimal selection, the present invention select input vector using the Forward-backward algorithm and minimum angle regression algorithm of current comparative maturity
In preferably input data be used as the input of model；On the basis of original ELM models, add multiple response sparse regression and calculate
Method and one by one extraction method crop hidden layer node useless or that effect is seldom, and use three kinds of neuronal activation letters of mixing
Number, so that the network established has more robustness and generalization；The present invention sufficiently considers different inputs to forecast model
Otherness, devise a kind of forecast model that each group of input data is dynamically updated using KALE adaptive algorithms, greatly
The precision for improving prediction.
Therefore, the present invention draws the Accurate Model of Volterra series and improved optimal beta pruning extreme learning machine
(OPELM, Optimally-Pruned Extreme Learning Machine) trains pace of learning comparatively fast and extensive, robust
A kind of the advantages of functional, it is proposed that IGBT method for predicting residual useful life based on VKOPP algorithms.Pass through accelerated life test
Characterization parameter of the collection emitter-base bandgap grading saturation voltage drop as IGBT module failure procedure is obtained, it is pre- with data-driven proposed by the present invention to it
Survey method is predicted, the results showed that, this method has good estimated performance.
Brief description of the drawings
Fig. 1 is a kind of embodiment flow chart of IGBT method for predicting residual useful life of the present invention；
Fig. 2 is the experiment initial data of 6 groups of IGBT collection emitter-base bandgap grading saturation voltage drops acquired in accelerated life test；
Fig. 3 is a kind of embodiment prediction result figure of IGBT method for predicting residual useful life of the present invention, wherein actual number
According to (dotted portion), prediction data (bold portion), APT represents acceptable performance threshold, i.e., failure threshold is (wherein, every
Group IGBT experimental datas are concentrated, and preceding 2/3rds data are used as training set, and rear 1/3rd data are used as test set).
Embodiment
The embodiment of the present invention is described below in conjunction with the accompanying drawings, so as to those skilled in the art preferably
Understand the present invention.Requiring particular attention is that in the following description, when known function and the detailed description of design perhaps
When can desalinate the main contents of the present invention, these descriptions will be ignored herein.
The present invention does not have to carry out model construction to specific circuit element, but is based on circuit electric network theory, utilizes electricity
The parameter measurability of network, it is combined and is designed using theoretical calculation and actual measurement.
Fig. 1 is a kind of embodiment flow chart of IGBT method for predicting residual useful life of the present invention.
In the present embodiment, as shown in figure 1, IGBT method for predicting residual useful life of the present invention can be divided into four-stage：
The acquisition of input data and processing stage, the initial phase of forecast model, saturation voltage drop data prediction stage and IGBT are remaining
The life prediction stage.Each stage is described in detail separately below：
(1), the acquisition of input data and processing stage
S101：By accelerated life test, IGBT collection emitter-base bandgap grading saturation voltage drop aging data time serieses are obtained, obtain data
Collect DS={ x (1), x (2) ..., x (D) }, wherein D is obtained data amount check；
S102：The data set DS obtained for step S101, first, determine that it is optimal embedding according to the method for differential entropy rate
Enter dimension d and delay time T；Secondly, framing reconstruct is carried out to data set DS using the function windowize in Matlab, and
Each frame data are respectively mapped to d dimensional feature spaces, the p in a manner of Volterra series (p >=1) rank is blocked, and is the letter of algorithm
Clean property, the present invention uses p=2, so as to obtain n (n=D-d τ) individual data frame { (Xt,Yt) (t=1,2 ..., n) to be used as training
Input data and the target output of model, wherein：
Xt=1, x (t), x (t+ τ) ..., x (t+ (d-1) τ),
x2(t), x (t) x (t+ τ) ... x (t) x (t+ (d-1) τ),
x2(t+ τ), x (t+ τ) x (t+2 τ) ..., x (t+ τ) x (t+ (d-1) τ),
x2(t+2 τ), x (t+2 τ) x (t+3 τ) ..., x (t+2 τ) x (t+ (d-1) τ)
...,
x2(t+ (d-2) τ), x (t+ (d-2) x (t+ (d-1) τ)
x2(t+(d-1)τ)}；
Yt=x (t+d τ)
Its element number is (d+1) (d+2)/2, target output data Yt=x (t+d τ) (t=1,2 ..., n)；
S103：The every frame input data X obtained using FB algorithms or LARS algorithms to step S102tMake optimal selection, select
Go out and export Y with targettThe higher variable of correlation, note selection after input data for：
Xt'={ 1, x (t+c1τ),x(t+c2τ),…,x(t+cbτ),…,x(t+ceτ)x(t+cfτ) ... },
Wherein：cb,ce,cf∈ 0,1 ..., d-1 }, and ce≤cf。Xt' dimension size (element number) be m, then might as well
Remember Xt'=[xt,1,xt,2,…,xt,m]T(t=1,2 ..., n).
2nd, the initial phase of forecast model
S104：One extreme learning machine (ELM) model for containing N (N ＜ D) individual hidden layer node of initialization.Input layer
Data are the X that step S103 is obtainedt' (t=1,2 ..., n).
In section [- 1,1] interior random initializtion ELM input weight coefficient matrixWith threshold coefficient matrixAccording to initializationWithCalculate hidden layer output matrixWherein
S105：By the hidden layer output matrix obtained in step S104And the corresponding target output of its in step S102
Yt(t=1,2 ..., n), using MRSR algorithms come vectorialElement be that hidden layer node is rearranged, note sequence after
Hidden layer node is：Wherein, 1≤j of subscripti≤ N is the hidden node sequence number before sequence,
1≤i of subscript≤N is the hidden node precedence sequence number after sequence；According to the order sequenced, l hidden layer node, wherein l before extraction
＜ N.
(3) the saturation voltage drop data prediction stage
S106：The hidden layer node input weights and threshold parameter matrix retained step S105 are updated to w=respectively
(ws,i)l×mWith θ=θs(s=1,2 ..., l；I=1,2 ..., m), and the new hidden layer of the network calculations after beta pruning is exported
MatrixWherein
S107：It is as follows using the network output weights after solution beta pruning and prediction result, its process：
1) predicts the D+1 data, that is, when predicting x (D+1), the first of forecast model is understood according to step S102 and S103
It is X to begin to inputt, now, t=D+1-d τ；After selection is inputted, note input is Xt' (t=D+1-d τ), then when this can be calculated
The network hidden layer output vector at quarter
2) calculates htThe matrix H obtained with step S106=[h1,h2,…,hj,…,hn] in each vectorial hj(j=1,
2 ..., n) Euclidean distance S；
3) is to h in StWith H each hj(j=1,2 ..., n) distance value is ranked up, and is found out from H closest to htZ
(z >=l) individual vector forms new matrixAnd corresponding target outputSo as to
Calculate corresponding output weights
T=D+1-d τ moment, the one-step prediction value of forecast model are obtained by the output weights：
4th, the IGBT predicting residual useful lifes stage
S108：Whether the predicted value that judgment step S107 is obtained reaches the performance failure threshold value A PT of IGBT module, if do not had
Have and reach, first, the input data frame of IGBT forecast models is updated using the method filled vacancies in the proper order of metabolism, predicted value that will be new
It is added to the afterbody of initial data frame, while removes the header data value of initial data frame, so ensures new data frame length
It is consistent with initial data frame.Then in repeat step S107 1), 2), 3), the network hidden layer for constantly updating the corresponding moment is defeated
Outgoing vector htWeights γ is exported with it.
If the predicted value of D+W (W >=1) individual dataIt is equal to or over its failure threshold APT, then record should
Critical point D+W.Then prediction, which obtains also W cycle count period IGBT pipe, will be in failure state.
In order to illustrate the technique effect of the present invention, 6 groups of IGBT accelerated life tests are completed altogether to verify.Fig. 2 is real
6 groups of acquired IGBT collection emitter-base bandgap grading saturation voltage drop aging datas time serieses are tested as experiment initial data EX_1~6.
The initial data shown in Fig. 2 is observed, it can be found that its data does not only exist bad point, also with redundancy and higher-dimension
The characteristics of, if being directly applied to life prediction, forecasting efficiency can be greatly reduced.Therefore, according to the actual needs and
The characteristic of problem, in the present embodiment, then conversion is compressed to the initial data extracted, obtained in lower dimensional space optimal
Feature, in favor of more accurately realizing IGBT life prediction.Fig. 3 (dotted portion) then depicts the data and curves after processing.
In the parameter setting of forecast model, when choosing optimal Embedded dimensions and delay using the method for differential entropy rate
Between；The nearest-neighbors number for setting model is that the exponent number of 100, Volterra series is three kind nerves of 2, the OPELM using mixing
First excitation function, and the number of hidden layer neuron are arranged to 100.
Under the conditions of above-mentioned parameter, it is the estimated performance for preferably verifying method proposed by the invention, then utilizes below
Different pieces of information number has carried out forecast analysis to IGBT residual life, is concentrated in every group of IGBT experimental data, preceding 2/3rds
Data be used as training set, rear 1/3rd data are used as test set.(note：Here training data points/instruction and hereafter
Practice the initial data that collection refers to the input of tectonic model and target output in step S102, test set then refers to step S107 and afterwards
Data to be predicted in step.)
Table 1 is prediction result of the present invention in the case where testing (1).
Table 1
When predicting IGBT residual life with IGBT method for predicting residual useful life proposed by the present invention as can be seen from Table 1,
Its relative error is no more than 1%.And it can also obtain preferable prediction result when using a small amount of data in forecast model.
From figure 3, it can be seen that IGBT method for predicting residual useful life of the present invention not only can be very good to predict IGBT experiment numbers
According to overall trend, can also relatively accurately predict its remaining lifetime value.
Although the illustrative embodiment of the present invention is described above, in order to the technology of the art
Personnel understand the present invention, it should be apparent that the invention is not restricted to the scope of embodiment, to the common skill of the art
For art personnel, if various change in the spirit and scope of the present invention that appended claim limits and determines, these
Change is it will be apparent that all utilize the innovation and creation of present inventive concept in the row of protection.
Claims (1)
1. a kind of IGBT method for predicting residual useful life, it is characterised in that comprise the following steps：
(1), by accelerated life test, IGBT collection emitter-base bandgap grading saturation voltage drop aging data time serieses is obtained, obtain data set DS
={ x (1), x (2) ..., x (D) }, wherein D is obtained data amount check；
(2), first, for data set DS, optimal Embedded dimensions d and delay time T are determined according to the method for differential entropy rate；So
Afterwards, framing reconstruct is carried out to data set DS, and each frame data is respectively mapped to d dimensional feature spaces, so as to obtain n data frame
{(Xt,Yt), wherein, n=D-d τ, t=1,2 ..., n, exported using the input data as training pattern and target, wherein：
Xt=1, x (t), x (t+ τ) ..., x (t+ (d-1) τ),
x2(t),x(t)x(t+τ),…x(t)x(t+(d-1)τ),
x2(t+τ),x(t+τ)x(t+2τ),…,x(t+τ)x(t+(d-1)τ),
x2(t+2τ),x(t+2τ)x(t+3τ),…,x(t+2τ)x(t+(d-1)τ)
...,
x2(t+(d-2)τ),x(t+(d-2)x(t+(d-1)τ)
x2(t+(d-1)τ)}；
Yt=x (t+d τ)
(3), the every frame input data X obtained using Forward-backward algorithm or minimum angle regression algorithm to step (2)tMake optimal choosing
Select, select and export Y with corresponding targettM higher input data of correlation, note selection after every frame input data for：
X′t={ 1, x (t+c1τ),x(t+c2τ),…,x(t+cbτ),…,x(t+ceτ)x(t+cfτ)}；cb,ce,cf∈{0,
1 ..., d-1 }, and ce≤cf；
Then per frame input data X 'tDimension size be that element number is m, can be expressed as again：
X′t=[xt,1,xt,2,…,xt,m]T, wherein, t=1,2 ..., n；
(4) an extreme learning machine model for containing N number of hidden layer node, is initialized, wherein, N ＜ D；The data of input layer are
The input data X ' that step (3) obtainst, wherein, t=1,2 ..., n；
In the input weight coefficient matrix of section [- 1,1] interior random initializtion extreme learning machineAnd threshold coefficient
MatrixWherein s=1,2 ..., N；I=1,2 ..., m；
According to the input weight coefficient matrix of initializationWith threshold coefficient matrixCalculate hidden layer output matrixWherein network hidden layer output vectorFor：
G represents excitation function；
(5), the vector obtained according to step (4)And the corresponding target output Y obtained in step (2)t, it is dilute using multiple response
Regression algorithm is dredged to vectorElement be that hidden layer node is rearranged, note sequence after hidden layer node be：
Wherein, 1≤j of subscriptk≤ N be sequence before hidden layer node sequence number, subscript 1≤
K≤N is the hidden layer node precedence sequence number after sequence；According to the order sequenced, before extractionl Individual hidden layer node, whereinl ＜
N；
(6), the input weight coefficient and threshold coefficient according to corresponding to step (5) extracts hidden layer node, input weight coefficient with
Threshold coefficient matrix is updated to w=(w respectivelys,i)l×mWith θ=θsS=1,2 ...,l ；I=1,2 ..., m, and be to extraction
After beta pruning, hidden layer output matrix H=[h are recalculated1,h2,…,hj,…,hn], wherein vector
(7) the network output weights after beta pruning, are solved and prediction result, its process are as follows：
7.1) the D+1 data, is predicted, i.e. during prediction data x (D+1), according at the beginning of forecast model knowable to step (2) and (3)
Beginning input data frame is Xt, now, t=D+1-d τ；After selection is inputted, input data frame is X ' after note selectiont, t=D+1-d
τ, then the network hidden layer output vector at the moment can be calculated i.e. t=D+1-d τ moment
7.2), calculating network hidden layer output vector htThe hidden layer output matrix H=[h obtained with step (6)1,h2,…,
hj,…,hn] in each vectorial hj, j=1,2 ..., n Euclidean distance S；
7.3), to network hidden layer output vector h in Euclidean distance StWith hidden layer output matrix H each vectorial hj, j=1,
2 ..., n distance value are ranked up, and are found out from hidden layer output matrix H closest to network hidden layer output vector htZ, z
>=l vector forms new matrixAnd corresponding target outputSo as to calculate
Corresponding output weights
T=D+1-d τ moment, the predicted value of forecast model are obtained by the output weights：
(8), the predicted value that judgment step (7) obtainsWhether IGBT performance failure threshold value, if reached, the prediction are reached
It is that D+1 records as critical point to be worth position, stops prediction, if be not reaching to, by predicted valueIt is added to IGBT collection emitter-base bandgap gradings
Saturation voltage drop aging data time series, data set DS={ x (2), x (3) ..., x (D+1) } is obtained, then according to step (1)
The method of~(7) obtains predicted valueAnd judge whether the performance failure threshold value for reaching IGBT, if it is not, continuously adding
To IGBT collection emitter-base bandgap grading saturation voltage drop aging data time serieses, data set DS={ x (3), x (4) ..., x (D+2) } is obtained, so
Repeat, recorded untill reaching, and using the predicted value position D+W as critical point, wherein, W is predicted valueReach IGBT
Performance failure threshold value prediction number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510396832.4A CN104978459B (en) | 2015-07-08 | 2015-07-08 | A kind of IGBT method for predicting residual useful life |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510396832.4A CN104978459B (en) | 2015-07-08 | 2015-07-08 | A kind of IGBT method for predicting residual useful life |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104978459A CN104978459A (en) | 2015-10-14 |
| CN104978459B true CN104978459B (en) | 2017-11-21 |
Family
ID=54274962
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510396832.4A Active CN104978459B (en) | 2015-07-08 | 2015-07-08 | A kind of IGBT method for predicting residual useful life |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104978459B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105486992B (en) * | 2015-11-05 | 2019-03-12 | 中车株洲电力机车研究所有限公司 | A kind of online health controller and method of insulated gate bipolar transistor |
| CN105320845A (en) * | 2015-11-26 | 2016-02-10 | 电子科技大学 | Time sequence forecast method based on quantum gravity algorithm |
| CN105550397B (en) * | 2015-12-03 | 2018-07-20 | 三峡大学 | A kind of IGBT module state evaluating method based on damage voltage |
| CN107861040A (en) * | 2016-09-22 | 2018-03-30 | 北京航空航天大学 | A kind of IGBT intermittent life test methods based on simulation modeling and short-term test |
| CN106528987B (en) * | 2016-11-03 | 2019-05-03 | 河北工业大学 | A kind of IGBT module accumulated damage for electric vehicle calculates and life-span prediction method |
| CN108169650B (en) * | 2016-12-06 | 2020-04-14 | 深圳市蓝海华腾技术股份有限公司 | Method and device for detecting whether service life of IGBT reaches standard |
| CN108804764A (en) * | 2018-05-07 | 2018-11-13 | 电子科技大学 | A kind of aging of lithium battery trend forecasting method based on extreme learning machine |
| CN109101738B (en) * | 2018-08-24 | 2022-11-15 | 河北工业大学 | IGBT module aging degree evaluation method |
| CN109840357B (en) * | 2019-01-08 | 2022-03-22 | 广东电网有限责任公司广州供电局 | Method and device for determining fatigue life of transistor module and computer equipment |
| CN112213660A (en) * | 2020-08-25 | 2021-01-12 | 广西电网有限责任公司南宁供电局 | Method for predicting residual life of power electronic device in UPS system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198223A (en) * | 2013-04-12 | 2013-07-10 | 电子科技大学 | Method for predicting real-time reliability of electronic products |
| CN103207362A (en) * | 2012-01-11 | 2013-07-17 | Abb研究有限公司 | System and method for monitoring in real time the operating state of an IGBT device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5343901B2 (en) * | 2010-03-17 | 2013-11-13 | 富士電機株式会社 | Power cycle life prediction method, life prediction device, and semiconductor device including the life prediction device |
| US9194376B2 (en) * | 2011-05-24 | 2015-11-24 | General Electric Company | System and method for estimating remaining life for a device |
-
2015
- 2015-07-08 CN CN201510396832.4A patent/CN104978459B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103207362A (en) * | 2012-01-11 | 2013-07-17 | Abb研究有限公司 | System and method for monitoring in real time the operating state of an IGBT device |
| CN103198223A (en) * | 2013-04-12 | 2013-07-10 | 电子科技大学 | Method for predicting real-time reliability of electronic products |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104978459A (en) | 2015-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104978459B (en) | A kind of IGBT method for predicting residual useful life | |
| Yuan et al. | Parameter extraction of solar cell models using chaotic asexual reproduction optimization | |
| CN109272146A (en) | A kind of Forecasting Flood method corrected based on deep learning model and BP neural network | |
| CN100367620C (en) | Power network topology error identification method based on mixed state estimation | |
| CN106295911B (en) | A kind of grid branch parameter evaluation method based on chromatographic assays | |
| CN103795057B (en) | Based on the power distribution network topology numbering generation method of search in distribution power system load flow calculation | |
| CN103268366A (en) | Combined wind power prediction method suitable for distributed wind power plant | |
| CN108021773B (en) | DSS database-based distributed hydrological model multi-field secondary flood parameter calibration method | |
| CN103995982B (en) | A kind of probability load flow calculation method for considering unit random fault | |
| CN104331572A (en) | Wind power plant reliability modeling method considering correlation between air speed and fault of wind turbine generator | |
| CN106022581B (en) | Based on geometry optimization-minimum variance method selective sampling Monte Carlo Model in Reliability Evaluation of Power Systems method | |
| US20240054267A1 (en) | Method for planning a layout of a renewable energy site | |
| CN110515931A (en) | A kind of capacitance type equipment failure prediction method based on random forests algorithm | |
| CN107368451A (en) | Circuit state space equation computational methods, device and system | |
| CN107966600A (en) | A kind of electricity anti-theft system and its electricity anti-theft method based on deep learning algorithm | |
| CN106786602B (en) | A kind of distribution power system load flow calculation method | |
| CN109285219A (en) | A kind of grid type hydrological model grid calculation order encoding method based on DEM | |
| CN105956252A (en) | Generative deep belief network-based multi-scale forecast modeling method for ultra-short-term wind speed | |
| CN104850918B (en) | A kind of node load Forecasting Methodology counted and power network topology constrains | |
| CN109447512B (en) | Large power grid reliability assessment method based on uniform design | |
| CN108665068A (en) | The improved adaptive GA-IAGA of water distribution hydraulic model automatic Check problem | |
| CN116720791B (en) | Distributed photovoltaic power generation quality management method and system | |
| CN116316617B (en) | Multi-station intelligent fusion new energy generation power region prediction method and system | |
| CN106295175B (en) | Station meteorological data missing value fill method based on svd algorithm | |
| CN105184383B (en) | City moving emergency power supply optimal scheduling method based on intelligent optimization method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |