Summary of the invention
For solving the deficiency that existing intelligent system exists, the present invention proposes a kind of new technical scheme, that is: three-dimensional Dynamic Uncertain cause-and-effect diagram (Cubic Dynamic Uncertain Causality Graph is called for short Cubic_DG) method for diagnosing faults.The technical scheme using the present invention to propose, by computer disposal, dynamically can generate three-dimensional DUCG (Dynamic Uncertain cause-and-effect diagram) online, and utilize this figure to carry out the fault diagnosis of objective system.Cubic_DG method proposed by the invention had both proposed a kind of based on original DUCG and dynamic evidence E (t
m) by computing machine automatically and dynamically generate Cubic_DG (t
m) method, it is also proposed based on Cubic_DG (t
m) the method for automatically being carried out fault diagnosis by computing machine.The present invention can be performed by the computing machine that at least one processor is housed, and this computing machine comprises further: input media, for the signal that input fault diagnosis is relevant; Output unit, for exporting generated three-dimensional DUCG figure, to carry out fault diagnosis.The Cubic_DG technology that application the present invention proposes, can process the dynamic fault diagnosis problem with degenerative objective system more rationally and effectively.Present patent application is the later patents application of having authorized Chinese invention patent ZL 2,006 8 0055266.X and U.S. patent Nos US 8255353 B2, is further developing above-mentioned granted patent technology.
Except term explained above, other relational language of explained later.
Similar with above-mentioned patent, in the present invention, so-called cause-effect relationship (causality), refers to the relation of logical relation between reason (cause) between anything and result (consequence) or effect (effect) or formal available causal logic relationship expression.Such as jet chimney leaks and vapor pressure will be caused to reduce and steam generator water level reduction, and leakage is reason, and it is result that pressure and water level reduce.
So-called objective system, comprises the industrial systems such as power plant, chemical plant, naval vessel, aerospace vehicle and other are various once occur abnormal symptom, need computer to diagnose its reason thus to take to successfully manage the system of measure.
So-called variable, expresses the event variable of proposition (proposition), exactly as temperature, pressure, flow, speed, frequency, content, proportion etc., its each state, as normal, higher, on the low side etc., be a value of variable, be also referred to as a state of variable.Causal variable V
irepresent, outcome variable X
nrepresent, no matter V
ior X
nwhether state itself discrete, be all discrete by its Unified Expression or fuzzy discrete state to adopt unified method to process, by V
iand X
ndifferent conditions Unified Expression be V
ijand X
nk, wherein i with n is the label distinguishing different variable, j with k is the label distinguishing different discrete or fuzzy discrete state, works as V
ior X
nduring for continuous variable, its any value e
ior e
nbe under the jurisdiction of V respectively
ijor X
nkdegree of membership m
ij(e
i) or m
nk(e
n) meet
or
v
ijand X
nktreated by as event, i.e. V
ijrepresent variable V
ibe in this event of j state, X
nkrepresent variable X
nbe in this event of k state, and if j ≠ j ', k ≠ k ', then V
ijwith V
ij 'mutual exclusion, X
nkwith X
nk 'mutual exclusion, if i ≠ i ', then B
ijand B
i ' j 'separate random occurrence, its probability of happening b
ij≡ Pr{B
ijbe known parameters, meet
Above-mentioned V ∈ { X, B, D, G}.X
nbeing outcome variable, also can be causal variable, Ke Yiyong
or
or other graphical symbol represents; B
ican only be causal variable, Ke Yiyong
or
or other graphical symbol represents; D
ix
idefault causal variable (or unknown cause variable), only have one for genuine state, Ke Yiyong
or other graphical symbol represents; G
ibe logic gate variable, express the cause-effect relationship between the various logic combination of its input variable and sub-variable, Ke Yiyong
or other graphical symbol represents, and use
or input variable is connected with logic gate by other graphical symbol, the particular content of logic gate is by logic gate instruction card LGS
i(Logic Gate Specification) illustrates:
j |
G
ij |
0 |
Logical expression 1 |
1 |
Logical expression 2 |
…… |
…… |
m |
Logical expression m |
Such as, " logical expression 1 "=X
1,1+ X
1,2b
4,2.Comma is for distinguishing variable label and state number.
As patent documentation (ZL 2,006 8 0055266.X) is recorded, actuation variable (Functional Variable) F
n; iexpress causal variable V
iwith outcome variable X
nbetween cause-effect relationship, F
nk; iexpress causal variable V
iwith result event X
nkbetween cause-effect relationship, F
n; Ijexpress reason event V
ijwith outcome variable X
nbetween cause-effect relationship, F
nk; Ijexpress reason event V
ijwith result event X
nkbetween cause-effect relationship, and can to represent with directed arc → or other graphical symbol, to point to result from reason.F
nk;ij≡(r
n;i/r
n)A
nk;ij。Wherein r
n; i>0 is causal variable V
iwith outcome variable X
nbetween strength of association (relationship intensity), r
n≡ ∑
ir
n; i, A
nk; Ijfor the V when not considering other causal variable
ijcause X
nkthere is this random occurrence, a
nk; Ij≡ Pr{A
nk; Ij, r
n; iand a
nk; Ijcan be the function of time, meet
f
nk; Ij≡ Pr{F
nk; Ij}=(r
n; i/ r
n) a
nk; Ij.F
nk; Ijv
ijto X
nkthe contribution margin of probability, meet
Above-mentioned actuation variable F
n; ican be conditioning variable, use dotted line directed arc
or other graphical symbol represents.Conditioning variable expresses its causal variable V
iwith outcome variable X
nbetween be conditioning relation, namely according to conditional event Z
n; iwhether meet to judge actuation variable F
n; iwhether set up.Such as Z
n; i=X
1,2, work as X
1,2for true time, Z
n; isatisfied, F
n; iestablishment, this conditioning variable
become actuation variable →; Work as X
1,2during non-real, Z
n; idissatisfied, F
n; ibe false, this conditioning variable
deleted.
An illustration of the DUCG built in the manner described above as shown in Figure 1.This figure corresponds to the Fig. 7 in above-mentioned patent specification, and be the DUCG graphic knowledge storehouse about nuclear power station secondary loop fault diagnosis, its variable description is see this patent specification, and two-way arc wherein represents two rightabout directed arcs.
Record according to the embodiment in above-mentioned patent documentation, t
1moment receives evidence
wherein E ' (t) represents abnormal evidence, E ' (t in this example
1)=X
6,2, E " and (t) represent normal evidence, in this example
in this instructions, state 0 is normal condition, state 1 is on the low side or the abnormality of two-value state variable, state 2 for higher, state 3 be low state, state 4 is high high state.Variable states can also be low amplitude vibrations, middle width vibrates and significantly vibrate, steadily rise, accelerate rising, steadily decline, accelerate lower degradation.According to the method that above-mentioned patent documentation is recorded, original DUCG figure (Fig. 1) can be that Fig. 2 (corresponding to the Figure 25-1 in above-mentioned patent documentation, has filled the X missed by abbreviation
3,0).Except subscript, can also represent state by color, such as green representative is normal, and yellow represents higher, and blueness represents on the low side, and grey represents the abnormality etc. of two-valued variable.
It is important to note that when objective system is the procedures system such as power plant, chemical plant, category-B type event exist just because of and non-ly just therefore to divide.So-called initial event, refers to that event that will directly cause objective system state to change occurs for it; So-called non-initial event, when referring to that certain initial event occurs, need certain parts or subsystem responses, and these parts or subsystem correctly can not respond, causes system to occur the event of special state.Non-initial event may occur already, but does not show before initial event occurs.The two or more independently simultaneous probability of initial event is high-order a small amount of, negligible.Have just because of and non-initial event exists time, non-initial event variable or event can be used
or
or other graphical symbol represents.At this moment the graphical symbol of single frame represents initial event variable or initial event.
For simplicity, below the state (that is event) of variable and variable is referred to as node.
Technical scheme of the present invention is as follows:
The structure that § 1, a kind of computing machine by being equipped with at least one processor perform is called as three-dimensional Dynamic Uncertain cause-and-effect diagram Cubic_DG (t
m) intelligent system and method in order to carry out objective system dynamic fault diagnosis, the method comprises the steps: that computing machine continues to scan the objective system signal by batch dynamically input, when detecting that the 1st batch exists abnormal signal, is t
1in the moment, detecting that the 2nd batch when being different from the abnormal signal of the 1st batch, is t
2moment, etc.Usually, when detect m criticize be different from m-1 criticize abnormal signal (being the normal signal be different from before this during m=1) time, be t
mmoment, t
mthe evidence in moment is E (t
m)=∏
je
j(t
m), wherein E
j(t
m) be the jth evidence that m criticizes in evidence, (1) works as t
m=t
1time, based on t
1evidence E (the t that moment obtains
1)=∏
je
j(t
1), make m=1, adopt the original DUCG figure of following rules reduction, the DUCG seal after the abbreviation obtained is Slice_DG (t
1), be also designated as Cubic_DG (t
1), i.e. Slice_DG (t
1)=Cubic_DG (t
1):
Rule 1: as E (t
m) display Z
n; ido not meet, by F
n; idelete from DUCG, as E (t
m) display Z
n; imeet, condition directed arc F
n; ibecome common directed arc F
n; i, figure is presented as
become →.
Rule 2: as E (t
m) display V
ij(V ∈ B, X}) be true, but V
ijbut not X
nreason event, by directed arc F
n; Ijdelete from DUCG.
Rule 3: as E (t
m) display X
nkbe true, but X
nkcan not by V
iany state of (V ∈ { B, X, G, D}) causes, by directed arc F
nk; idelete from DUCG.
Rule 4: as E (t
m) { this node and input directed arc thereof are deleted by B, X} type node Status unknown and no-output directed arc from DUCG in display.
Rule 5: as E (t
m) display X
n0be true, and X
n0with abnormal evidence E ' (t) without any connected relation, by X
n0delete from DUCG.
Rule 6: as E (t
m) show one group of Status unknown node except by X
n0, otherwise not with X
nkbe connected (k ≠ 0), this group Status unknown node and the directed arc that is attached thereto are deleted from DUCG.
Rule 7: because any reason causes G
ido not export, by G
iand input directed arc
delete from DUCG; As m=1, because any reason causes G
ido not input directed arc, by G
iand output directed arc is deleted from DUCG.
Rule 8: when directed arc does not have reason node or node of coming to nothing, this directed arc is deleted from DUCG.
Rule 9: as existence one group node and directed arc and E (t
m) in the node that relates to without connected relation, this group node and directed arc are deleted from DUCG.
Rule 10: if 1. E (t
m) show or cause some V
ij, but V
nstatus unknown, { B, X, G}, n ≠ i, 2. according to DUCG, V for V ∈
nbe only V
ijfather node or child node, and F
nk; Ij≠ 0 or F
ij; Nk≠ 0, F
nk '; Ij=0 or F
ij; Nk '=0, k ≠ k ', then by the V in DUCG
nchange V into
nk.
Rule 11: above-mentioned rule can be used alone according to any order, conbined usage, reuse.
It should be noted that, above-mentioned rule is equally applicable to the technical scheme described in above-mentioned patent documentation.
(2) as m>1, based on t
mevidence E (the t that moment obtains
m)=∏
je
j(t
m), the original DUCG figure of abbreviation, is designated as Slice_DG (t
m), comprise the steps: that (a) uses above-mentioned rule to carry out abbreviation to original DUCG figure, but except following (b) and (c) described situation; (b) in execution step (a), if Slice_DG (t
m-1) in there is X
n0, and E (t
m) in also comprise X
n0, retain nodes X
n0; (c) in execution step (a), if X
nkdo not input directed arc, and k ≠ 0, retain nodes X
nk.
(3) by Slice_DG (t
m) and Cubic_DG (t
m-1) carry out the connection of crossing time sheet, generate three-dimensional DUCG and scheme Cubic_DG (t
m), comprise the steps:
A () is by Slice_DG (t
m) and Cubic_DG (t
m-1) category-B type node that interior joint label is identical overlaps as Cubic_DG (t after getting and occuring simultaneously
m-1) in node.
B () is by Cubic_DG (t
m-1) in node as being different from Slice_DG (t
m) in node treat, scheme based on original DUCG, find Cubic_DG (t
m-1) in whether there is Slice_DG (t
m) in result node X
n(t
m) or X
nk(t
m) reason node V
i(t
m ') or V
ij(t
m '), { B, X, G}, if existed, couple together result node and reason node with the corresponding F type directed arc in original DUCG figure, and figure is presented as use directed arc → by V for m ' <m and V ∈
i(t
m ') or V
ij(t
m ') and X
n(t
m) or X
nk(t
m) couple together, point to X from V.
C () is by Slice_DG (t
m) and Cubic_DG (t
m-1) middle t
m-1the undirected arc of X type node----that nodal scheme in timeslice and state are all identical or other symbol couples together, deletes Slice_DG (t simultaneously
m) in this X type node with undirected arc and Cubic_DG (t
m-1) in identical and its upstream also identical input directed arc of coupling together of identical node.So-called upstream is exactly reason direction.
If d () Cubic_DG is (t
m-1) in there is X
n0(t
m '), m ' <m, simultaneously Slice_DG (t
m) in there is X
nk(t
m), k ≠ 0, deletes Cubic_DG (t
m-1) in X
n0(t
m ') and the Status unknown X of upstream and G type node and the directed arc that is attached thereto.
If e () Cubic_DG is (t
m-1) in the X of existence the unknown
n(t
m '), m ' <m, simultaneously Slice_DG (t
m) in there is X
nk(t
m), delete Cubic_DG (t
m-1) in X
n(t
m ') with the directed arc be attached thereto, to delete with deleted output directed arc be the node uniquely inputting directed arc and the node be connected with undirected arc with this node thereof simultaneously.
If f () Slice_DG is (t
m) in exist not have input G type node, at Cubic_DG (t
m-1) in searching whether there is the input node relevant with its output directed arc of this G type node, if found, using the input of input node as this G type node, figure is presented as use
input node is connected to G type node.
(g) according to above-mentioned steps obtain by Slice_DG (t
m) and Cubic_DG (t
m-1) figure that links together is called as t
mthe three-dimensional DUCG figure in moment, is designated as Cubic_DG (t
m), which includes from t
1to t
meach timeslice (time slice) in node and cross over and do not cross over the F type directed arc of timeslice.Export Cubic_DG (t
m), to provide current possible fault, and these faults cause objective system that current abnormal causal logic relation occurs, thus carry out fault diagnosis, to take reasonable measure elimination fault.
§ 2, one are based on Cubic_DG (t
m), carried out the method for objective system fault diagnosis by the computing machine that at least one processor is housed, comprise the steps:
(1) to Cubic_DG (t
m) in node and F type directed arc mark the timestamp of its place timeslice, wherein, V
nor V
nk, { B, X, D, G} are labeled as V to V ∈
i(t
p) or V
ij(t
p), F
nk; Ijbe labeled as F
nk; Ij(t
p, t
q), wherein p ∈ 1,2 ..., m} marks X
nkplace timeslice, q ∈ 1,2 ..., m} marks V
ijplace timeslice, q≤p.
(2) to t
m≤ t≤t
m+1, Cubic_DG (t will be contained in
m) in Status unknown event and combination be defined as assumed fault event H
kj∈ S
h(t), wherein,
for Cubic_DG (t
m) in evidence, abnormality evidence E ' (t) and normal condition evidence E can be decomposed into " (t); it can be used as the abnormality evidence E ' in above-mentioned patent documentation claim 10 and normal condition evidence E respectively " treat, according to the method described in above-mentioned patent documentation, logic expansion is carried out to E ' (t), obtain all possible breakdown hypothesis events under E (t) or event sets H
kj∈ S
h(t).E ' can be called as common abnormal evidence, does not namely consider the abnormal evidence of time attribute, is made up of the logical and of several abnormal signal events, such as E '=X
1,2x
2,3x
3,1, etc.; E " common normal evidence can be called as, namely do not consider the normal evidence of time attribute, such as E "=X
4,0x
5,0x
7,0, etc.So-called E ' (t) be used as E ' treat, exactly by E ' (t) not evidence in the same time regard different abnormal evidences as.Such as E ' (t)=X
1,2(t
1) X
2,3(t
2) X
3,1(t
2) X
1,2(t
3) X
3,3(t
4) be exactly be made up of, although X 5 different evidences
1,2(t
1) and X
1,2(t
3) be that identical variable states repeats, X
3,1(t
2) and X
3,3(t
4) be identical variable states contradiction, but because they are in different time points respectively, therefore treated by as different evidences.That is, time point is also used for distinguishing different evidences by one of attribute as evidence, and in effect, these different evidences are relations of logical and; Similarly, if E " (t)=X
4,0(t
1) X
5,0(t
2) X
4,0(t
3), X
4,0(t
1) and X
4,0(t
3) be also two different evidences.
(3) by Cubic_DG (t
m) as the DUCG in above-mentioned patent documentation, will
as the E in above-mentioned patent documentation, the method recorded according to above-mentioned patent documentation calculates H
kj∈ S
hthe conditional probability Pr{H of (t)
kj| E (t) }=Pr{H
kje (t) }/Pr{E (t) }, wherein relate to and launch set event E (t) and H
kje (t), no longer comprises X and G type node in the expression formula after expansion.So-called common evidence E is exactly E=E ' E ", wherein E ' and E " implication as above-mentioned.Obviously, E (t)=E ' (t) E " (t).
(4) in the deployment step of above-mentioned (3), the different conditions of identical variable is in not mutual exclusion in the same time, but in mutual exclusion in the same time mutually, the equal state of identical variable not independent in the same time, but is absorbing mutually in the same time mutually.Such as, as k ≠ k ', X
nk(t
1) X
nk '(t
2) ≠ 0, X
nk(t
2) X
nk '(t
2)=0, F
nk; Ij(t
2, t
1) F
nk '; I ' j '(t
3, t
2) ≠ 0, F
nk; Ij(t
2, t
1) F
nk '; I ' j '(t
2, t
2)=0, X
nk(t
2) X
nk(t
3) ≠ X
nk(t
2) or ≠ X
nk(t
3), X
nk(t
2) X
nk(t
2)=X
nk(t
2).
(5) in the process performing (3), the undirected arc of the connection same node point of different time sheet is crossed over according to the process of extension mode.So-called extend, be exactly the node node of comparatively early timeslice and input directed arc thereof being substituted into more late timeslice, the timestamp of coupled F type directed arc also does corresponding conversion, remerges identical directed arc.
(6) based on the logical expression that above-mentioned steps is launched, the event in formula or event variable probability of occurrence corresponding in original DUCG or event variable probability matrix are substituted, calculate H
kj∈ S
hthe state probability of (t)
with sequence probability
(7) export above-mentioned (6) described state probability and sequence probability, so that user understands probability of happening and the sequence thereof of current possible breakdown, thus take reasonable measure and eliminate fault as early as possible.
§ 3, when application is industrial process systems (process system), the technical scheme that can realize described in § 1 and § 2 according to following step:
(1) original DUCG can be decomposed according to initial event variable, namely retain the initial event variable B in original DUCG one by one
i, delete remaining initial event variable B
i ', i ' ≠ i, then make m=1 abbreviation only comprise B according to the described rule of § 1 (1)
ioriginal DUCG figure, obtain carrying out the multiple DUCG figure after decomposing abbreviation according to initial event variable, be designated as DUCG (B
i).
(2) DUCG (B is used one by one
i) as original DUCG, according to step described in § 1, build and B
icorresponding Slice_DG (B
i, t
m) and Cubic_DG (B
i, t
m).
(3) step when (2) can not realize constructing complete Cubic_DG (B
i, t
m) time, namely cannot construct Cubic_DG (B
i, t
m) explain all abnormal nodes in E (t) time, by Cubic_DG (B
i, t
m) delete.
(4) remaining Cubic_DG (B is exported
i, t
m), so that user understands and currently has which possible breakdown, and these faults cause industrial process systems to occur abnormal causal logic relation.
(5) if remaining Cubic_DG (B
i, t
m) more than one, calculate its Pr{E (t) respectively }, be designated as ζ (B
i, t
m), wherein relate to the calculating launching E (t) identical with step described in claim 2.
(6) if remaining Cubic_DG (B
i, t
m) more than one, calculate Cubic_DG (B respectively
i, t
m) weight factor
(7) Cubic_DG (B is calculated according to the step of claim 2
i, t
m) in comprise by initial event variable B
iin initial event and possible non-initial event form possible breakdown hypothesis H
kjstate probability
and revise by weight factor:
Calculate sequence probability again
(8) export (7) gained state probability and sequence probability, so that user understands probability of happening and the sequence thereof of current possible breakdown, instruct user to take reasonable measure as early as possible and fix a breakdown.
The technical scheme of § 3 as shown in Figure 3.
Embodiment
Method of the present invention is described in detail below in conjunction with accompanying drawing.
Example 1 nuclear power station secondary loop dynamic fault diagnosis
The original DUCG figure of nuclear power station secondary loop as shown in Figure 1, its variable declaration is see patent documentation (ZL 2006 80055266.X) table 1.Wherein X
n0both without input, also no-output.All r
n; i=1.Because parameter is too many, when not affecting explanation technical solution of the present invention, only provide part b and a type parameter herein:
Wherein "-" expression is not paid close attention to, and is equivalent to " 0 ".Every parameter is "-" or " 0 " place, and cause-effect relationship does not exist, i.e. corresponding F
nk; Ij=0 or A
nk; Ij=0.
Because this example is procedures system, according to § 3 (1), decompose the DUCG (B after abbreviation
102) as shown in Figure 4 (other decomposes abbreviation figure slightly).Below by DUCG (B
102) treat as the original DUCG figure in patent documentation (ZL 2,006 8 0055266.X).
1. t
1moment receives evidence
according to § 1 (1) described step, abbreviation is carried out to Fig. 4, the Slice_DG (t after abbreviation
1)=Cubic_DG (B
102, t
1) as shown in Figure 2.
Based on Cubic_DG (t
1), according to § 2 (1), (2) described step, launch E ' (t)=X
6,2(t
1), E ' (t)=F can be obtained
6,2; 102,1(t
1, t
1) B
102,1.After removing F categorical variable, S can be obtained
h(t)={ B
102,1.
According to the described step of § 1 (1) to other DUCG (B
i), i ≠ 102, carry out abbreviation, and launch E ' (t)=X according to § 2 (1), (2) described step
6,2(t
1), acquired results is S
h(t)=0 (empty set).According to § 3 (3), these Cubic_DG (B
i, t
m), i ≠ 102, are deleted.
According to § 2 (3), (6) described step, and notice § 2 (4), the situation of (5) do not exist herein, calculates H
kj=B
102,1state probability and sequence probability: launch E (t)=X
6,2x
3,0x
12,0x
15,0x
23,0in X
6,2, obtain
E(t)=F
6,2;102,1(t
1,t
1)B
102,1X
3,0X
12,0X
15,0X
23,0
B
102,1E(t)=B
102,1F
6,2;102,1(t
1,t
1)B
102,1X
3,0X
12,0X
15,0X
23,0
=F
6,2;102,1(t
1,t
1)B
102,1X
3,0X
12,0X
15,0X
23,0=E(t)
Wherein X
3,0, X
12,0, X
15,0, X
23,0should according to X
n0=(1-∑
k ≠ 0x
nk) mode launch.But in this example, due to B
102,1e (t)=E (t), no longer launches for the sake of simplicity.So have
2. t
2the evidence that moment receives is
According to § 1 (2), based on the Slice_DG (B after Fig. 4 abbreviation
102, t
2) as shown in Figure 5.
According to § 1 (3) (a), by the Slice_DG (B in Fig. 5
102, t
2) with Fig. 2 in Slice_DG (B
102, t
1) identical category-B type variable get occur simultaneously after overlap, obtain result shown in Fig. 6.
The situation of § 1 (3) (b) does not exist in this example.
According to § 1 (3) (c), by Slice_DG (B
102, t
2) and Slice_DG (B
102, t
1) in X
6,2use undirected arc----to couple together, delete Slice_DG (B simultaneously
102, t
2) in X
6,2node with Slice_DG (B
102, t
1) in X
6,2the input directed arc F that node is identical
6,2; 102,1, obtain the result of Fig. 7.
According to § 1 (3) (d), (g), by t in Fig. 5
1the X in moment
12,0, X
15,0, X
23,0and input directed arc F
12,0; 102,1, F
15,0; 102,1and F
23,0; 102,1leave out, obtain the result shown in Fig. 8, be Cubic_DG (B
102, t
2).
The situation of § 1 (3) (e), (f) does not exist in this example.
Based on Cubic_DG (B
102, t
2), can obtain the evidence wherein comprised is:
E(t)=E
6(t
1)E
1(t
2)E
2(t
2)E
3(t
2)E
12(t
2)E
13(t
2)E
15(t
2)E
23(t
2)E
26(t
2)=X
1,2X
2,2X
3,2X
6,2X
12,2X
13,2X
15,0X
23,2X
26,1
According to step described in § 2, each evidence carries out logic expansion according to Fig. 8 and substitutes into above-mentioned evidence expression formula carrying out logical and respectively, and the logic expansion result of E (t) is as follows, wherein for the sake of simplicity, timestamp is omitted (will embody in example 2), and according to F
nk; Ij=(r
n; i/ r
n) A
nk; Ijlaunch, calculate weight coefficient.
E(t)=(1/24)A
13,2;1,2A
1,2;23,2A
2,2;23,2A
3,2;23,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/48)A
13,2;1,2A
1,2;23,2A
2,2;23,2A
3,2;13,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/48)A
13,2;1,2A
1,2;23,2A
2,2;23,2A
3,2;13,2A
13,2;2,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/48)A
13,2;1,2A
1,2;23,2A
2,2;13,2A
3,2;23,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/96)A
13,2;1,2A
1,2;23,2A
2,2;13,2A
3,2;13,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/96)A
13,2;1,2A
1,2;23,2A
2,2;13,2A
3,2;13,2A
13,2;2,2A
2,2;23,2X
23,2A
23,2;102,1A
26,1;23,2A
6,2;102,1
A
12,2;102,1A
15,0;102,1B
102,1
+(1/24)A
13,2;2,2A
1,2;23,2A
2,2;23,2A
3,2;23,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+(1/48)A
13,2;2,2A
1,2;23,2A
2,2;23,2A
3,2;13,2A
23,2;102,1A
26,1;23,2A
6,2;102,1A
12,2;102,1A
15,0;102,1B
102,1
+ ... (totally 81)
Carry out logic expansion to E (t) in Fig. 8, delete after category-A type event and weight coefficient remerge similar events, each affair logic remaining in expression formula and item are possible hypothesis event.In this example, this possible hypothesis event is B
102,1, i.e. S
h(t)={ B
102,1, because above-mentioned every middle category-B type event all only has B
102,1.
Due to S
han event B is only had in (t)
102,1, according to the described step of § 2 (3)-(6), B
102,1state and sequence probability must be
3. t
3the evidence that moment receives is:
According to § 1 (2), the Slice_DG (B after abbreviation
102, t
3) as shown in Figure 9.
According to § 1 (3) (a), by the Slice_DG (B in Fig. 9
102, t
3) with Fig. 8 in Cubic_DG (B
102, t
2) identical category-B type variable get occur simultaneously after overlap, obtain result shown in Figure 10.
The situation of § 1 (3) (b) does not exist herein.
According to § 1 (3) (c), by Slice_DG (B
102, t
3) and Cubic_DG (B
102, t
2) in t
2x in timeslice
6,2and X
26,1use undirected arc----to couple together respectively, delete Slice_DG (B simultaneously
102, t
3) in X
6,2with Cubic_DG (B
102, t
2) in t
2the X in moment
6,2the input directed arc F of the same node point be connected by undirected arc
6,2; 102,1, obtain the result of Figure 11.
According to § 1 (3) (d), (g), by t in Figure 11
2the X in moment
15,0and input directed arc F
15,0; 102,1delete, obtain the result shown in Figure 12, be Cubic_DG (B
102,t
3).
The situation of § 1 (3) (e), (f) does not exist in this example.
Based on Cubic_DG (B
102,t
3), be similar to the calculating 2. of this example, still can obtain S
h(t)={ B
102,1, and the result of calculation of its state probability and sequence probability constant (process slightly).
4. t
4the evidence that moment receives is:
According to § 1 (2), the Slice_DG (B after abbreviation
102,t
3) as shown in figure 13.
According to § 1 (3) (a), by the Slice_DG (B in Figure 13
102,t
4) with Figure 12 in Cubic_DG (B
102,t
3) identical category-B type variable get occur simultaneously after overlap, obtain result shown in Figure 14.
The situation of § 1 (3) (b) does not exist herein.
According to § 1 (3) (c), by Slice_DG (B
102,t
4) and Cubic_DG (B
102,t
3) in t
3x in timeslice
6,2, X
15,2and X
26,1use undirected arc----to couple together respectively, delete Slice_DG (B simultaneously
102,t
4) in X
6,2and X
15,2with Cubic_DG (B
102,t
3) in t
3the X in moment
6,2and X
15,2the input directed arc F of the same node point be connected by undirected arc
6,2; 102,1and F
15,2; 102,1obtain the result of Figure 15.
The situation of § 1 (3) (e), (f) does not exist in this instance, and thus result shown in Figure 15 is Cubic_DG (B
102,t
4).
Based on Cubic_DG (B
102,t
4), be similar to the calculating 2. of this example, can S be obtained
h(t)={ B
102,1, and the result of calculation of its state probability and sequence probability constant (process slightly).
5. t
5the evidence that moment receives is:
According to § 1 (2), the Slice_DG (t after abbreviation
5) as shown in figure 16.
According to § 1 (3) (a), by the Slice_DG (B in Figure 16
102,t
5) with Figure 15 in Cubic_DG (B
102,t
4) identical category-B type variable get occur simultaneously after overlap, obtain result shown in Figure 17.
The situation of § 1 (3) (b) does not exist herein.
According to § 1 (3) (c), by Slice_DG (B
102,t
5) and Cubic_DG (B
102,t
4) in t
4x in timeslice
1,2, X
2,1, X
3,1, X
6,2, X
13,1, X
15,2and X
26,1use undirected arc----to couple together respectively, delete Slice_DG (B simultaneously
102,t
5) in X
1,2, X
2,1, X
3,1, X
6,2, X
13,1, X
15,2and X
26,1with Cubic_DG (B
102,t
4) in t
4the X in moment
1,2, X
2,1, X
3,1, X
6,2, X
13,1, X
15,2and X
26,1the input directed arc F of the same node point be connected by undirected arc
1,1; 13,1, F
2,1; 13,1, F
3,1; 13,1, F
13,1; 1,1, F
13,1; 2,1, F
13,1; 3,1, F
23,1; 1,1, F
23,1; 2,1, F
23,1; 3,1, F
6,2; 102,1and F
15,2; 102,1obtain the result of Figure 18.
The situation of § 1 (3) (e), (f) does not exist in this instance, and thus result shown in Figure 18 is Cubic_DG (B
102,t
5).
Based on Cubic_DG (B
102,t
5), be similar to the calculating 2. of this example, can S be obtained
h(t)={ B
102,1, and the result of calculation of its state probability and sequence probability constant (process slightly).The fault that Figure 18 reflects 5 timeslices occurs and evolution, clearly presents vapor (steam) temperature X
23with reactor capability X
26between positive feedback and negative-feedback signal oscillatory condition, original DUCG figure and reasoning process do not use aforementioned condition 1 and condition 2, do not use Treatment yet.
Example 2
Figure 19 is an original DUCG figure for fault diagnosis, and described object is a procedures system, wherein all r
n; i=1, conditioning event Z
4; 2=X
3,2.Interrelated logic door and parameter values as follows:
In this example, we are first according to the three-dimensional DUCG figure of method construct described in § 1, below again according to procedures system first because of and non-initial event split, observe multiple the DUCG figure split according to initial event are put in a figure, essence be equivalent to application § 3 described in step.
Below be described in four moment in succession receive new abnormal evidence successively after three-dimensional DUCG figure growth course, mainly with t
3moment is example, describes the technical scheme how applying the present invention's proposition in detail and carries out fault diagnosis.
1. t
1moment
T
1the evidence that moment receives is
Slice_DG (t is obtained according to § 1 (1) abbreviation
1) (i.e. Cubic_DG (t
1)) as shown in figure 20.Wherein B
1,1and B
1,2all X can be caused
1,1and X
2,1.
2. t
2moment
T
2the evidence that moment receives is
According to the Slice_DG (t that § 1 (2) abbreviation obtains
2) as shown in figure 21.
According to § 1 (3), Slice_DG (t
2) and Cubic_DG (t
1) connect, obtain Cubic_DG (t
2) as shown in figure 22.3. t
3moment
T
3the evidence that moment receives is
Slice_DG (t is obtained according to § 1 (2) abbreviation
3) as shown in figure 23.
According to § 1 (3), by the Slice_DG (t in Figure 23
3) with Figure 22 in Cubic_DG (t
2) connect, obtain Cubic_DG (t
3) as shown in figure 24.Wherein, application § 1 (3) (e), the X in Figure 22
14and X
5,0deleted.X is embodied in Figure 24
4,2(t
2) event initiation X
15,1(t
3), X
15,1(t
3) merge again non-initial event B
4,1acting in conjunction causes X
4state fall back to signal X on the low side
4,1(t
3) concussion process.
According to the described initial event decomposing scheme of § 3 (1), and according to step described in § 2, Cubic_DG (B can be obtained
1, t
3) proof logic launch result be:
E(t)={(1/3)[A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,1(t
1;t
1)
A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
2)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,1(t
1;t
1)
+A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,1(t
1;t
1)A
8,1;6,2(t
1;t
1)
A
6,2;9,1(t
2;t
2)A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
1)A
6,2;2,1(t
1;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,1(t
1;t
1)
+A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,1(t
1;t
1)A
9,1;8,1(t
2;t
2)
A
8,1;6,2(t
2;t
1)A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,1(t
1;t
1)]B
1,1B
4,1
+(1/3)[A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,2(t
1;t
1)
A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
2)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,2(t
1;t
1)
+A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,2(t
1;t
1)A
8,1;6,2(t
1;t
1)
A
6,2;9,1(t
2;t
2)A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
1)A
6,2;2,1(t
1;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,2(t
1;t
1)
+A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;1,2(t
1;t
1)A
9,1;8,1(t
2;t
2)
A
8,1;6,2(t
2;t
1)A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1)A
2,1;1,2(t
1;t
1)]B
1,2B
4,1}
A
13,0;13D(t
2;t
2)
Substitute the A in above formula with a, substitute the B in above formula with b, and bring the value of a and b type parameter into above formula and calculate, can obtain
Pr{E(t)}=0.00000008535;
Cubic_DG (B
2, t
3) in evidence logic launch result be:
E(t)=(1/6)[(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))
(A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)(A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
2)A
6,2;2,1(t
2;t
1))A
6,4;2,1(t
3;t
1)A
2,1;2,2(t
1;t
1))
(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
+(1/6)[(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))
(A
8,1;6,2(t
1;t
1)(A
6,2;9,1(t
2;t
2)A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
1))A
6,2;2,1(t
1;t
1)A
6,4;2,1(t
3;t
1)A
2,1;2,2(t
1;t
1))
(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
+(1/6)[(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;3,2(t
2;t
2)A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))
((A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
1)A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1))A
2,1;2,2(t
1;t
1))
(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
+(1/6)[(A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))(A
8,1;6,2(t
1;t
1)A
6,2;2,1(t
1;t
1)(A
9,1;8,1(t
2;t
2)
A
8,1;6,2(t
2;t
2)A
6,2;2,1(t
2;t
1))A
6,4;2,1(t
3;t
1)A
2,1;2,2(t
1;t
1))(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)
A
4,2;2,2(t
2;t
1))(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
+(1/6)[(A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))(A
8,1;6,2(t
1;t
1)(A
6,2;9,1(t
2;t
2)A
9,1;8,1(t
2;t
2)
A
8,1;6,2(t
2;t
1))A
6,2;2,1(t
1;t
1)A
6,4;2,1(t
3;t
1)A
2,1;2,2(t
1;t
1))(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)
A
4,2;2,2(t
2;t
1))(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
+(1/6)[(A
3,2;1,1(t
2;t
2)A
13,0;13D(t
2;t
2)A
1,1;2,2(t
1;t
1))((A
9,1;8,1(t
2;t
2)A
8,1;6,2(t
2;t
1)A
8,1;6,2(t
1;t
1))
A
6,2;2,1(t
1;t
1)A
6,2;2,1(t
2;t
1)A
6,4;2,1(t
3;t
1))A
2,1;2,2(t
1;t
1)(A
4,1;3,1(t
3;t
3)A
15,1;4,2(t
3;t
2)A
4,2;2,2(t
2;t
1))
(A
5,0;14(t
1;t
1)A
14;2,2(t
1;t
1))A
14,0;2,2(t
3;t
1)]B
2,2B
4,1
Substitute the A in above formula with a, substitute the B in above formula with b, and bring the value of a and b type parameter into above formula and calculate, can obtain
Pr{E(t)}=0.000000004344。
According to step shown in § 3 (4), (5), Cubic_DG (B can be obtained
1, t
3) and Cubic_DG (B
2, t
3) corresponding weight coefficient is respectively: ξ (B
1, t
2)=0.9516, ξ (B
2, t
2)=0.04844.
Finally calculate to obtain t
3moment fault hypothesis space is S
h(t)={ B
1,1b
4,1, B
1,2b
4,1, B
2,2b
4,1.According to step shown in § 2, each hypothesis event H can be obtained
kj∈ S
ht the state probability of () is:
Sequence probability is:
Wherein, due to B
1,1with B
1,2mutual exclusion, must have B
1,1b
4,1with B
1,2b
4,1state probability sum equal 1.Again due to ξ (B
1, t
2) and ξ (B
2, t
2) be normalizing, so state probability equals the probability that sorts in this example.
4. t
4moment
T
4the evidence that moment receives is
Abbreviation obtains Slice_DG (t
4) as shown in figure 25.
Be similar to this example 3., Slice_DG (t
4) and Cubic_DG (t
3) connect obtain Cubic_DG (t
4) and obtain Figure 26 (Cubic_DG (B respectively after decomposing by initial event
1, t
4)) and Figure 27 (Cubic_DG (B
2, t
4)).
As seen from Figure 26, at Cubic_DG (B
1, t
4) in, abnormal evidence X
14,1(t
4) can not get any explanation, therefore, according to § 3 (3), B
1corresponding three-dimensional DUCG schemes Cubic_DG (B
1, t
4) deleted.
T
4the effective three-dimensional DUCG figure that moment finally obtains only has one, as shown in figure 27.Calculate according to § 2 (4), (5) described step, ξ (B must be had
2, t
4)=1.S can be obtained according to the described step of § 2 (2)
h(t)={ B
2,2b
4,1.
Due to S
ht () only comprises an assumed fault, calculate, must have according to the described step in § 2 (3)-(6)
namely diagnostic result is diagnosed as by initial event B
2,2with non-initial event B
4,1concurrent, jointly form fault rootstock.
industrial applicability
The present invention can industrially apply.Its effect provides the valuable fresh informations such as objective system health monitoring, failure prediction, diagnosis, prediction and decision support for people, to improve security and the availability of objective system.