CN103617350B - A kind of rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence - Google Patents

Abstract

The present invention relates to a kind of rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence, belong to rotating machinery failure monitoring and diagnostic techniques field.Diagnostic evidence when equipment runs is obtained by evidence acquisition methods, recursively use conditioning linear weighted function evidence fusion rule, realize the diagnostic evidence after a upper moment is updated by current time diagnostic evidence to be updated, thus obtain the diagnostic evidence after current time updates.Under certain decision rule, the diagnostic evidence after updating carry out fault decision-making.Based on current, history and the reliability of future time instance diagnostic evidence, determine the smoothing weights of evidence during linear fusion updates, diagnostic evidence after this makes obtained current time update dynamically contains the useful diagnostic message of history and future time instance, so that based on the diagnosis decision-making made of diagnostic evidence after updating, than based on do not do decision-making that the diagnostic evidence updated makes the most accurate with reliably.

Description

A kind of rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence

Technical field

The present invention relates to a kind of rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence, belong to rotation Mechanical equipment fault detection and diagnostic techniques field.

Background technology

On-line fault diagnosis technology is to ensure rotating machinery safety in production and the powerful measure of Effec-tive Function, but mesh In the case of before, this technology is the most at the early-stage, is also faced with lot of challenges in implementation process.Due to answering of fault mode and feature thereof Polygamy and multiformity, the detection of traditional the most not competent fault of information processing method based on single-sensor and diagnosis, want Realize real-time diagnosis and improve fault diagnosis rate, using multisensor to increase amount of diagnostic information imperative.In information fusion Evidence theory and method with it probabilistic representing, measure and the aspect such as fusion has the advantage that, become fault diagnosis neck A kind of method being progressively taken seriously in territory.In existing various fusion diagnosis methods based on evidence theory, if document is " based on mould Stick with paste fault characteristic information random set metric fusion diagnosis method, electronics and information journal " in proposition diagnostic evidence obtain Take and fusion method, owing to judging fault only with Current Diagnostic evidence, do not account for Current Diagnostic evidence and history and future Variation tendency between diagnostic evidence and relation, will make final diagnosis decision-making lack enough accuracys and reliability.

Summary of the invention

It is an object of the invention to, proposed a kind of smooth the rotating machinery fault of renewal based on diagnostic evidence and examines Disconnected method, is updated current time diagnostic evidence, history and future time instance diagnostic evidence merging, utilize obtain current time Carve the diagnostic evidence after updating and make diagnosis decision-making, so that diagnostic result is more accurately with reliable.

The present invention propose a kind of rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence, including with Under each step:

(1) failure collection of rotating machinery is set as Θ={ F₀,F₁,…,F_j,…,F_N, F_jRepresent rotating machinery to set Standby jth kind fault (j=0,1 ..., N), then have N+1 kind fault.

(2) by diagnostic evidence generate method, can the kth moment (k=1,2,3 ...), it is thus achieved that rotating machinery Diagnostic evidence is E_k=(m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ)), wherein m_k(F_j) represent in the k moment, right The reliability assignment that jth kind fault occurs, m_k(Θ) the reliability assignment to failure collection, then E are represented_kConstitute for these reliability assignment A reliability adele, and have 1-(m_k(F₀)+m_k(F₁)+…+m_k(F_j)+…+m_k(F_N))=m_k(Θ)。

(3) diagnostic evidence obtained based on step (2), by linear weighted function diagnostic evidence fusion rule, uses the kth moment Diagnostic evidence carries out smooth renewal to historical diagnostic evidence, thus obtain the k moment update after diagnostic evidence E_1:k=(m_1:k (F₀),m_1:k(F₁),…,m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), wherein 1:k represents E_1:kBe merge from 1 to k the moment own Diagnostic evidence obtain, specifically comprise the following steps that

(3-1) as k=1, the diagnostic evidence after renewal is

E_1:1=E₁

That is the diagnostic evidence after updating is the diagnostic evidence of this moment acquisition；

(3-2) the diagnostic evidence vector E when k >=2, after renewal_1:k, its each element value is given by with following formula (1) and (2) Go out

m_1:k(A)=α_km_1:k-1(A)+β_km_k(A|B) A,B∈Θ (1)

m_1:k(Θ)=1-Σ_A∈Θm_1:k(A) (2)

Wherein, the m in formula (1)_1:k-1(A) represent the k-1 moment update after diagnostic evidence E_1:k-1Reliability assignment to fault A； m_k(A | B) represents the conditioning reliability assignment about fault A that the kth moment obtains, and works as A=F_jTime, if the diagnosis card in k moment According to E_k=(m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ) in), m_k(F_j) more than other all of m_k(F₀),m_k (F₁),…,m_k(F_j-1),…,m_k(F_j+1),…,m_k(F_N), then B=F_j, m_k(A | B)=1；Otherwise, m_k(A | B)=0；

α_kAnd β_kFor linear fusion smoothing weights, solution procedure is as follows:

A () be diagnostic evidence E after the acquisition k-1 moment updates_1:k-1, k moment diagnostic evidence E_kWith k+1 moment diagnostic evidence E_k+1Afterwards, formula (3) is utilized to calculate vector E_1:k-1With vector E_kBetween distance d (E_1:k-1,E_k) it is

$d(E_{1:k-1},E_k)=\sqrt{\frac{1}{2}(E_{1:k-1}-E_k)\underset{=}{D}{(E_{1:k-1}-E_k)}^T}---\left(3\right)$

E in formula_1:k-1-E_kBeing the vector obtained after two vector corresponding elements subtract each other, T represents the transposition of vector,It it is one (N+2) matrix of × (N+2), its diagonal entry value is 1, and the element value of N+1 row is walked in the 1st of its N+2 row, And the element value that the 1st row of N+2 row arrange to N+1 is 1/ (N+1), other element values are 0;

In like manner, formula (4) vector E is obtained_1:k-1With vector E_k+1Between distance d (E_1:k-1,E_k+1) it is

$d(E_{1:k-1},E_{k+1})=\sqrt{\frac{1}{2}(E_{1:k-1}-E_{k+1})\underset{=}{D}{(E_{1:k-1}-E_{k+1})}^T}---\left(4\right)$

Vector E is obtained by formula (5)_kWith vector E_k+1Between distance d (E_k,E_k+1) it is

$d(E_k,E_{k+1})=\sqrt{\frac{1}{2}(E_k-E_{k+1})\underset{=}{D}{(E_k-E_{k+1})}^T}---\left(5\right)$

B () is tried to achieve distance d (E by above-mentioned steps (a)_1:k-1,E_k)、d(E_1:k-1,E_k+1) and d (E_k,E_k+1After), calculate the k moment E_1:k-1、E_kAnd E_k+1Similarity between any two:

Formula (6) is utilized to calculate vector E_1:k-1With vector E_kBetween similarity c (E_1:k-1,E_k) it is

c(E_1:k-1,E_k)=1-d (E_1:k-1,E_k) (6) similarity c (E_1:k-1,E_k) it is to weigh vector E_1:k-1With vector E_k Close degree, that is the degree that two evidences are consistent, and have c (E_1:k-1,E_k)=c (E_k,E_1:k-1), i.e. vector E_1:k-1With vector E_kSimilarity equal to vector E_kWith vector E_1:k-1Similarity；

In like manner, formula (7) vector E is obtained_1:k-1With vector E_k+1Between similarity c (E_1:k-1,E_k+1) it is

c(E_1:k-1,E_k+1)=1-d (E_1:k-1,E_k+1) (7)

Vector E is obtained by formula (8)_kWith vector E_k+1Between similarity c (E_k,E_k+1) it is

c(E_k,E_k+1)=1-d (E_k,E_k+1) (8)

C () obtains at k moment diagnostic evidence vector E according to above-mentioned steps (b)_1:k-1、E_kAnd E_k+1Similarity c two-by-two (E_1:k-1,E_k), c (E_1:k-1,E_k+1) and c (E_k,E_k+1After), calculate each evidence vector and supported by other two evidence vectors Support:

Formula (9) is utilized to calculate evidence vector E_1:k-1By evidence vector E_kAnd E_k+1Support s (the E supported_1:k-1) it is

s(E_1:k-1)=c (E_1:k-1,E_k)+c(E_1:k-1,E_k+1) (9)

Support s is the function of similarity measurement, represents the degree that this evidence is supported by other evidences, s (E_1:k-1) value The highest, then explanation evidence vector E_1:k-1With evidence vector E_kAnd E_k+1Between similarity the highest；

In like manner, formula (10) evidence vector E is calculated_kBy evidence vector E_1:k-1And E_k+1Support s (the E supported_k) it is

s(E_k)=c (E_1:k-1,E_k)+c(E_k,E_k+1) (10)

Evidence vector E is calculated by formula (11)_k+1By evidence vector E_1:k-1And E_kSupport s (the E supported_k+1) it is

s(E_k+1)=c (E_1:k-1,E_k+1)+c(E_k,E_k+1) (11)

D () obtains diagnostic evidence vector E successively based on step (c)_1:k-1、E_kAnd E_k+1Reliability K in the k moment: utilize public affairs Formula (12) calculates evidence vector E_1:k-1Reliability K (E in the k moment_1:k-1) it is

$K\left(E_{1:k-1}\right)=\frac{s\left(E_{1:k-1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(12\right)$

In like manner, formula (13) evidence vector E is calculated_kReliability K (E in the k moment_k) it is

$K\left(E_k\right)=\frac{s\left(E_k\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(13\right)$

Evidence vector E is calculated by formula (14)_k+1Reliability K (E in the k moment_k+1) it is

$K\left(E_{k+1}\right)=\frac{s\left(E_{k+1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(14\right)$

There is K (E_1:k-1)+K(E_k)+K(E_k+1)=1, reliability K of an evidence vector is the highest, illustrates that this evidence is demonstrate,proved with other According to similarity the highest, this evidence is the most reliable, and vice versa；

E E that () is tried to achieve based on above-mentioned steps (b)_1:k-1With E_k+1Between similarity c (E_1:k-1,E_k+1)、E_kWith E_k+1It Between similarity c (E_k,E_k+1), by judging that size therebetween determines α_k, β_kValue:

If c is (E_1:k-1,E_k+1)≥c(E_k,E_k+1), $\left\{\begin{array}{c}{\mathrm{\α}}_k=K\left(E_{1:k-1}\right)+K\left(E_{k+1}\right)\\ {\mathrm{\β}}_k=K\left(E_k\right)\end{array}\right.$

If c is (E_1:k-1,E_k+1) ＜ c (E_k,E_k+1), $\left\{\begin{array}{c}{\mathrm{\α}}_k=K\left(E_{1:k-1}\right)\\ {\mathrm{\β}}_k=K\left(E_k\right)+K\left(E_{k+1}\right)\end{array}\right.$

Try to achieve α_k, β_kValue after substituted in (1) formula of step (3-2), when can obtain each by recursive calculation Carve the diagnostic evidence after updating.

(4) according to above-mentioned steps (3) diagnostic evidence E after the renewal that the k moment obtains_1:k=(m_1:k(F₀),m_1:k(F₁),…, m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), the diagnosing malfunction to rotating machinery: if m_1:k(F_j) value more than its He is m_1:k(F₀),m_1:k(F₁),…,m_1:k(F_j-1),…,m_1:k(F_j+1),…,m_1:k(F_N), then judge fault F_jOccur.

The key technology of said method is: the linear fusion smoothing weights α solved in step (3-2)_k, β_k, consider Current, variation tendency between history and future time instance diagnostic evidence and relation, overcome conventional diagnostic evidence fusion method, The impact on current time diagnostic evidence of the diagnostic message of history and future time instance is not considered when making diagnosis decision-making, caused The diagnosis shortcoming that decision-making is inaccurate and reliability is low.

The rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence that the present invention proposes, can realize rotating The on-line fault diagnosis of plant equipment, effectively monitors the running status of equipment, can remind in time when there is emergency Operator carry out overhaul plan, promote reliability and the accuracy of diagnosis significantly.And what fault was developed by the present invention Trend is followed the tracks of, and fault also functions to certain predicting function to a certain extent.Do not send out in diagnostic evidence When giving birth to big fluctuation, the diagnostic evidence after renewal can be kept substantially consistent with not updating diagnostic evidence and be conducive to equipment Running status is made and being judged accurately；When diagnostic evidence occurs change drastically, the diagnostic evidence after renewal can reflect This change, and can quickly follow the tracks of the variation tendency of fault reliability, thus realize faster, more accurate diagnosis decision-making.According to this The program (translation and compiling environment LabVIEW, C++ etc.) of bright method establishment can be run on monitoring computer, and combination sensor, number According to hardware composition on-line monitoring systems such as harvesters, carry out detection and the diagnosis of real-time rotating machinery fault.

Accompanying drawing explanation

Fig. 1 is the FB(flow block) of the inventive method.

Rotor fault diagnosis system structure chart in the embodiment of Fig. 2 the inventive method.

Fig. 3 be the inventive method embodiment in rotor be in normal operation mode (F₀Each fault reliability assignment under) Trend graph (k=1,2 ..., 10 moment).

Fig. 4 be the inventive method embodiment in rotor in the k=7 moment by normal operation mode (F₀) gradually change For imbalance fault (F₁) time each fault reliability assignment trend graph (k=1,2 ..., 10 moment).

Fig. 5 be in the embodiment of the present invention rotor in the k=6 moment by normal operation mode (F₀) it is changed into suddenly injustice Weighing apparatus fault (F₁) time each fault reliability assignment trend graph (k=1,2 ..., 10 moment).

Fig. 6 is that in the embodiment of the present invention, rotor is in normal operation mode (F when starting₀), in the k=6 moment by dry Disturb and (show as imbalance fault F₁) again recover normal condition time each fault reliability assignment trend graph (k=1,2 ..., when 10 Carve).

Fig. 7 is that in the embodiment of the present invention, rotor is in normal operation mode (F when starting₀), in k=6 and the k=7 moment It is interfered respectively and (shows as imbalance fault F respectively₁With misalign fault F₂) time each fault reliability assignment trend graph (k= 1,2 ..., 10 moment).

Fig. 8 is that in the embodiment of the present invention, rotor is in normal operation mode (F when starting₀), it is subject to suddenly in the k=5 moment (imbalance fault F is shown as to interference₁) recover again normal condition, then, being interfered again in the k=7 moment, (performance misaligns Fault F₂) time each fault reliability assignment trend graph (k=1,2 ..., 10 moment).

Detailed description of the invention

The rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence that the present invention proposes, its FB(flow block) As it is shown in figure 1, include following steps:

(1) failure collection of rotating machinery is set as Θ={ F₀,F₁,…,F_j,…,F_N, F_jRepresent rotating machinery to set Standby jth kind fault (j=0,1 ..., N), then have N+1 kind fault；

(2) by diagnostic evidence generate method, can the kth moment (k=1,2,3 ...), it is thus achieved that rotating machinery Diagnostic evidence is E_k=(m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ)), wherein m_k(F_j) represent in the k moment, right The reliability assignment that jth kind fault occurs, m_k(Θ) the reliability assignment to failure collection, then E are represented_kConstitute for these reliability assignment A reliability adele, and have 1-(m_k(F₀)+m_k(F₁)+…+m_k(F_j)+…+m_k(F_N))=m_k(Θ)；

(3) diagnostic evidence obtained based on step (2), by linear weighted function diagnostic evidence fusion rule, uses the kth moment Diagnostic evidence carries out smooth renewal to historical diagnostic evidence, thus obtain the k moment update after diagnostic evidence E_1:k=(m_1:k (F₀),m_1:k(F₁),…,m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), wherein 1:k represents E_1:kBe merge from 1 to k the moment own Diagnostic evidence obtain, specifically comprise the following steps that

(3-1) as k=1, the diagnostic evidence after renewal is

E_1:1=E₁

That is the diagnostic evidence after updating is the diagnostic evidence of this moment acquisition；

(3-2) the diagnostic evidence vector E when k >=2, after renewal_1:k, its each element value is given by with following formula (1) and (2) Go out

m_1:k(A)=α_km_1:k-1(A)+β_km_k(A|B) A,B∈Θ (1)

m_1:k(Θ)=1-Σ_A∈Θm_1:k(A) (2)

Wherein, the m in formula (1)_1:k-1(A) represent the k-1 moment update after diagnostic evidence E_1:k-1Reliability assignment to fault A； m_k(A | B) represents the conditioning reliability assignment about fault A that the kth moment obtains, and works as A=F_jTime, if the diagnosis card in k moment According to E_k=(m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ) in), m_k(F_j) more than other all of m_k(F₀),m_k (F₁),…,m_k(F_j-1),…,m_k(F_j+1),…,m_k(F_N), then B=F_j, m_k(A | B)=1；Otherwise, m_k(A | B)=0；

α_kAnd β_kFor linear fusion smoothing weights, solution procedure is as follows:

A () be diagnostic evidence E after the acquisition k-1 moment updates_1:k-1, k moment diagnostic evidence E_kWith k+1 moment diagnostic evidence E_k+1Afterwards, formula (3) is utilized to calculate vector E_1:k-1With vector E_kBetween distance d (E_1:k-1,E_k) it is

$d(E_{1:k-1},E_k)=\sqrt{\frac{1}{2}(E_{1:k-1}-E_k)\underset{=}{D}{(E_{1:k-1}-E_k)}^T}---\left(3\right)$

E in formula_1:k-1-E_kBeing the vector obtained after two vector corresponding elements subtract each other, T represents the transposition of vector,It it is one (N+2) matrix of × (N+2), its diagonal entry value is 1, and the element value of N+1 row is walked in the 1st of its N+2 row, And the element value that the 1st row of N+2 row arrange to N+1 is 1/ (N+1), other element values are 0;

In like manner, formula (4) vector E is obtained_1:k-1With vector E_k+1Between distance d (E_1:k-1,E_k+1) it is

$d(E_{1:k-1},E_{k+1})=\sqrt{\frac{1}{2}(E_{1:k-1}-E_{k+1})\underset{=}{D}{(E_{1:k-1}-E_{k+1})}^T}---\left(4\right)$

Vector E is obtained by formula (5)_kWith vector E_k+1Between distance d (E_k,E_k+1) it is

$d(E_k,E_{k+1})=\sqrt{\frac{1}{2}(E_k-E_{k+1})\underset{=}{D}{(E_k-E_{k+1})}^T}---\left(5\right)$

In order to deepen the understanding to diagnostic evidence vector distance, citing here is illustrated: assume to obtain fault diagnosis Evidence E_k=(m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ)), k=1 is taken here, 2,3, N=2.Each moment diagnoses Reliability assignment corresponding to evidence is as shown in table 1, and wherein, k represents each moment, and m represents reliability assignment.

The diagnostic evidence that the table 1k=1,2,3 moment obtains

Obtained by step (3-1): the k=1 moment update after diagnostic evidence E_1:1For

E_1:1=E₁=(0.9 0.04 0.03 0.03)

Know according to step (a),It is the diagonal matrix of 4 × 4, is expressed as follows:

$\underset{=}{D}=\left[\begin{array}{cccc}1& 0& 0& \frac{1}{3}\\ 0& 1& 0& \frac{1}{3}\\ 0& 0& 1& \frac{1}{3}\\ \frac{1}{3}& \frac{1}{3}& \frac{1}{3}& 1\end{array}\right]$

MatrixMiddle element calculates and only listsIt is explained, ${\underset{=}{D}}_{\mathrm{2,4}}=D\left(\right\{F_0\},\mathrm{\Θ})=\left|F_0\right|/\left|\mathrm{\Θ}\right|=1/3$

Diagnostic evidence vector E can be obtained by table 1 and formula (3)_1:1With vector E₂Between distance d (E_1:1,E₂) it is

E_1:1-E₂=(0.05-0.02-0.02-0.01)

$d(E_{1:1},E_2)=\sqrt{\frac{1}{2}(E_{1:1}-E_2)\underset{=}{D}{(E_{1:1}-E_2)}^T}=0.0577$

In like manner, d (E can be obtained according to formula (4) and (5)_1:1,E₃) and d (E₂,E₃) it is d (E_1:1,E₃)=0.1871,

d(E₂,E₃)=0.1354

B () is tried to achieve distance d (E by above-mentioned steps (a)_1:k-1,E_k)、d(E_1:k-1,E_k+1) and d (E_k,E_k+1After), calculate the k moment E_1:k-1、E_kAnd E_k+1Similarity between any two:

Formula (6) is utilized to calculate vector E_1:k-1With vector E_kBetween similarity c (E_1:k-1,E_k) it is

c(E_1:k-1,E_k)=1-d (E_1:k-1,E_k) (6)

Similarity c (E_1:k-1,E_k) it is to weigh vector E_1:k-1With vector E_kClose degree, that is the journey that two evidences are consistent Degree, and have c (E_1:k-1,E_k)=c (E_k,E_1:k-1), i.e. vector E_1:k-1With vector E_kSimilarity equal to vector E_kWith vector E_1:k-1 Similarity；

In like manner, formula (7) vector E is obtained_1:k-1With vector E_k+1Between similarity c (E_1:k-1,E_k+1) it is

c(E_1:k-1,E_k+1)=1-d (E_1:k-1,E_k+1) (7)

Vector E is obtained by formula (8)_kWith vector E_k+1Between similarity c (E_k,E_k+1) it is

c(E_k,E_k+1)=1-d (E_k,E_k+1) (8)

C () obtains at k moment diagnostic evidence vector E according to above-mentioned steps (b)_1:k-1、E_kAnd E_k+1Similarity c two-by-two (E_1:k-1,E_k), c (E_1:k-1,E_k+1) and c (E_k,E_k+1After), calculate each evidence vector and supported by other two evidence vectors Support:

Formula (9) is utilized to calculate evidence vector E_1:k-1By evidence vector E_kAnd E_k+1Support s (the E supported_1:k-1) it is

s(E_1:k-1)=c (E_1:k-1,E_k)+c(E_1:k-1,E_k+1) (9)

Support s is the function of similarity measurement, represents the degree that this evidence is supported by other evidences, s (E_1:k-1) value The highest, then explanation evidence vector E_1:k-1With evidence vector E_kAnd E_k+1Between similarity the highest；

In like manner, formula (10) evidence vector E is calculated_kBy evidence vector E_1:k-1And E_k+1Support s (the E supported_k) it is

s(E_k)=c (E_1:k-1,E_k)+c(E_k,E_k+1) (10)

Evidence vector E is calculated by formula (11)_k+1By evidence vector E_1:k-1And E_kSupport s (the E supported_k+1) it is

s(E_k+1)=c (E_1:k-1,E_k+1)+c(E_k,E_k+1) (11)

D () obtains diagnostic evidence vector E successively based on step (c)_1:k-1、E_kAnd E_k+1Reliability K in the k moment: utilize public affairs Formula (12) calculates evidence vector E_1:k-1Reliability K (E in the k moment_1:k-1) it is

$K\left(E_{1:k-1}\right)=\frac{s\left(E_{1:k-1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(12\right)$

In like manner, formula (13) evidence vector E is calculated_kReliability K (E in the k moment_k) it is

$K\left(E_k\right)=\frac{s\left(E_k\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(13\right)$

Evidence vector E is calculated by formula (14)_k+1Reliability K (E in the k moment_k+1) it is

$K\left(E_{k+1}\right)=\frac{s\left(E_{k+1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(14\right)$

There is K (E_1:k-1)+K(E_k)+K(E_k+1)=1, reliability K of an evidence vector is the highest, illustrates that this evidence is demonstrate,proved with other According to similarity the highest, this evidence is the most reliable, and vice versa；

E E that () is tried to achieve based on above-mentioned steps (b)_1:k-1With E_k+1Between similarity c (E_1:k-1,E_k+1)、E_kWith E_k+1It Between similarity c (E_k,E_k+1), by judging that size therebetween determines α_k, β_kValue:

If c is (E_1:k-1,E_k+1)≥c(E_k,E_k+1), $\left\{\begin{array}{c}{\mathrm{\α}}_k=K\left(E_{1:k-1}\right)+K\left(E_{k+1}\right)\\ {\mathrm{\β}}_k=K\left(E_k\right)\end{array}\right.$

If c is (E_1:k-1,E_k+1) ＜ c (E_k,E_k+1), $\left\{\begin{array}{c}{\mathrm{\α}}_k=K\left(E_{1:k-1}\right)\\ {\mathrm{\β}}_k=K\left(E_k\right)+K\left(E_{k+1}\right)\end{array}\right.$

Try to achieve α_k, β_kValue after substituted in (1) formula of step (3-2), when can obtain each by recursive calculation Diagnostic evidence after the renewal carved；

In order to make it easy to understand, provide instantiation here, based on upper example, the renewal result of diagnostic evidence when seeking k=2:

Based on the distance between the diagnostic evidence vector that upper example is tried to achieve, formula (6) vector E can be obtained_1:1With vector E₂ Between similarity c (E_1:1,E₂)=1-d (E_1:1,E₂)=0.9423

In like manner, formula (7) and (8) c (E is obtained_1:1,E₃)=0.8129, c (E₂,E₃)=0.8646

Diagnostic evidence vector E can be calculated successively according to formula (9)～(14)_1:1、E₂And E₃Reliability in the k=2 moment It is respectively K (E_1:1)=0.3350, K (E₂)=0.3449, K (E₃)=0.3202

α is determined according to step (e)₂, β₂Value as follows:

Because c is (E_1:1,E₃) ＜ c (E₂,E₃), therefore $\left\{\begin{array}{c}{\mathrm{\α}}_1=K\left(E_{1:1}\right)=0.3350\\ {\mathrm{\β}}_2=K\left(E_2\right)+K\left(E_3\right)=0.6650\end{array}\right.$

By formula (1) and (2), the diagnostic evidence in k=2 moment is updated, because when the k=2 moment, m (F₀) value is Greatly, therefore when determining conditioning reliability assignment, B=F₀；Work as A=F₀Time, m₂(F₀|F₀)=1；Otherwise, m₂(A|F₀)=0(A=F₁、F₂、 Θ), result such as table 2 is updated

Diagnostic evidence after updating during table 2k=2 moment

(4) according to above-mentioned steps (3) diagnostic evidence E after the renewal that the k moment obtains_1:k=(m_1:k(F₀),m_1:k(F₁),…, m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), the diagnosing malfunction to rotating machinery: if m_1:k(F_j) value more than its He is m_1:k(F₀),m_1:k(F₁),…,m_1:k(F_j-1),…,m_1:k(F_j+1),…,m_1:k(F_N), then judge fault F_jOccur.

Below in conjunction with accompanying drawing, the embodiment of the inventive method it is discussed in detail:

The FB(flow block) of the inventive method is as it is shown in figure 1, core is: adjacent by solving history, current and future Distance between three moment diagnostic evidence vectors, obtains the similarity between vector two-by-two, then obtains linear weighted function diagnosis Linear fusion smoothing weights during evidence fusion, thus realize the renewal to historical diagnostic evidence, diagnosed accurately and reliably Result.

Below in conjunction with the most preferred embodiment of rotor fault diagnosis system in Fig. 2, the inventive method each is discussed in detail Individual step, and smooth the method for diagnosing faults of renewal based on diagnostic evidence be better than it by what the actual result checking present invention proposed His method.

1, rotor fault diagnosis system arranges example

The experimental facilities such as ZHS-2 type multi-function motor flexible rotor system in Fig. 2, by vibration displacement sensor and acceleration Degree sensor is respectively disposed on the both horizontally and vertically collection rotor oscillation signal of rotor supports seat, through HG-8902 vasculum Transmit a signal to computer, then utilize the HG-8902 data analysis software under Labview environment to obtain rotor oscillation and accelerate Degree frequency spectrum and time domain vibration displacement average amplitude are as fault characteristic signals.

2, rotor fault is arranged and the choosing of Fault characteristic parameters

On testing stand, rotor is provided with fault F respectively₀={ properly functioning }, F₁={ uneven }, F₂={ the most right In and F₃={ pedestal loosens }, then failure collection i.e. framework of identification is Θ={ F₀,F₁,F₂,F₃}.By the operation shape of monitoring device Condition, obtains the evidence in sensor the most continuous 10 moment.

In conjunction with six diagnostic tests (three kinds of situations) described in most preferred embodiment Fig. 3 to Fig. 8, by the inventive method and biography The linear combination method of system contrasts, and to highlight the advantage of the inventive method, makes a concrete analysis of as follows:

2-1) situation one: rotor kth (k=1,2 ..., 10) moment is always maintained at normal operating condition, time different Carve the diagnostic evidence (in Fig. 3, "--◇--" represents) obtained as shown in table 3:

Known by table 3, when equipment is properly functioning, when between variant moment diagnostic evidence vector, difference is inconspicuous, with this The trend graph of each fault reliability assignment that inventive method and linear combination method obtain is as shown in Figure 3: the most why do not have Provide m_k(Θ) trend graph, is because m_k(Θ) can become the least after merging to such an extent as to have little influence on fault Diagnosis and decision-making.

In this case, because system is always to maintain properly functioning, therefore m_k(F₀) remain maximum.Therefore, after renewal To F in diagnostic evidence₀Reliability assignment the most maximum, from Fig. 3 it is apparent that the F that calculates of the inventive method₀Reliability Assignment (in Fig. 3 "--◇--" represent), the reliability assignment (in Fig. 3, "--*--" represents) calculated than linear combination method is more Greatly, it is more beneficial for reliably judging that equipment is now in F₀, and pass through F₁、F₂、F₃Trend graph analysis, be not difficult to learn this Invention can also be effectively reduced the reliability assignment of other faults, it is to avoid is disturbed by these uncertain reliabilities in decision-making.

The diagnostic evidence that each moment of table 3 obtains

2-2) situation two: when rotor is transitioned into malfunction by normal operating condition, now transition state can be divided again Gradual change (Fig. 4) and sudden change (Fig. 5) two kinds of situations.

A () rotor is in normal operating condition in the k=1 moment to the k=7 moment, the most gradually break down F₁.Fig. 4 In give the diagnostic evidence updated before merging, and diagnostic evidence trend graph after the renewal that obtains of two kinds of methods.

B () rotor k=1 moment to the k=6 moment is in normal operating condition, the k=7 moment breaks down suddenly F₁Until The k=10 moment.Now the renewal result of diagnostic evidence is as shown in Figure 5:

From the situation of Fig. 4 fault reliability assignment generation gradual change it can be seen that in the k=8 moment, when fault F₁Reliability assignment When becoming larger, after the renewal that the inventive method is given in diagnostic evidence, to F₁Reliability assignment increase substantially (value 0.75) And after the renewal that linear combination method is given in diagnostic evidence, to F₁Reliability assignment only have 0.48, now the method is mistaken for F₀.As time goes on, the inventive method is to F₁Reliability assignment be always maintained at maximum, and linear combination method is to F₁Letter Degree assignment increasess slowly, and less with raw diagnostic evidence gap, illustrates that it updates the effect merged little.Believe from Fig. 5 fault The situation that degree assignment is undergone mutation is it can be seen that in the k=7 moment, raw diagnostic evidence is to F₁Reliability assignment increase suddenly, it After be always maintained at a high position, F is described₁Suddenly occur, to F in diagnostic evidence after the renewal that now the inventive method is given₁Reliability compose Value increase substantially (value 0.78) and after renewal that linear combination method is given in diagnostic evidence, to F₁Reliability assignment only 0.21, now the method is mistaken for F₀.As time goes on, the inventive method is to F₁Reliability assignment be always maintained at maximum, and Linear combination method is to F₁Reliability assignment increases slowly, and less with raw diagnostic evidence gap, illustrate that it updates and merge Effect is little.The reliability broken down, it can be seen that diagnostic evidence after the renewal that is given of the inventive method, is composed by Integrated comparative Result that value is given always above linear combination method and stable holding, be conducive to providing more structurally sound diagnostic result.

2-3) situation three: receive external interference in rotor normal course of operation the most in the same time, once disturb disappearance, Rotor returns to normal operating condition again, and now system mode can be subdivided into again three sub cases:

A () rotor is only interfered in the k=6 moment and (shows as imbalance fault F₁), the now renewal of diagnostic evidence Result is as shown in Figure 6:

B () rotor is continually subjected to interference in k=6 and the k=7 moment and (shows as fault F respectively₁And fault F₂), now examine The renewal result of disconnected evidence is as shown in Figure 7:

C () rotor, by interval disturbance, is interfered in the k=5 moment and (shows as fault F₁), it is subject to again in the k=7 moment (fault F is shown as to interference₂), now the renewal result of diagnostic evidence is as shown in Figure 8:

One preferable fault diagnosis system should have certain immunization, i.e. system to external interference and run Even if journey is disturbed by certain, also diagnostic result will not be caused ill effect.From Fig. 6, Fig. 7 and Fig. 8 it can be seen that this Invention and linear combination method can be cut down the interference of fault to a certain extent and (that is in interference, moment, interference institute table be occurred The reliability assignment revealing the malfunction come is not higher than F₀Reliability assignment), although when interference occur time linear combination to interference The assignment of the malfunction shown is slightly lower than the assignment that the inventive method provides, but after interference disappears, side of the present invention Method is to F₀Reliability assignment the most high, thus be more beneficial for the diagnosis of fault, while can making accurate decision-making, this Invent selection the most more preferable, more structurally sound.

Claims (1)

1. the rotating machinery method for diagnosing faults smoothing renewal based on diagnostic evidence, it is characterised in that the method includes Following steps:

Step (1) sets the failure collection of rotating machinery as Θ={ F₀,F₁,…,F_j,…,F_N, F_jRepresent rotating machinery to set Standby jth kind fault, j=0,1 ..., N, then have N+1 kind fault；

Step (2) generates method by diagnostic evidence, in the kth moment, it is thus achieved that the diagnostic evidence of rotating machinery is E_k= (m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ)), wherein, k=1,2,3 ...；m_k(F_j) represent in the k moment, right The reliability assignment that jth kind fault occurs, m_k(Θ) the reliability assignment to failure collection, then E are represented_kConstitute for these reliability assignment A reliability adele, and have 1-(m_k(F₀)+m_k(F₁)+…+m_k(F_j)+…+m_k(F_N))=m_k(Θ)；

The diagnostic evidence that step (3) obtains based on step (2), by linear weighted function diagnostic evidence fusion rule, uses the kth moment Diagnostic evidence carries out smooth renewal to historical diagnostic evidence, thus obtain the k moment update after diagnostic evidence E_1:k=(m_1:k (F₀),m_1:k(F₁),…,m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), wherein 1:k represents E_1:kBe merge from 1 to k the moment own Diagnostic evidence obtain, specifically comprise the following steps that

(3-1) as k=1, the diagnostic evidence after renewal is

E_1:1=E₁

That is the diagnostic evidence after updating is the diagnostic evidence of this moment acquisition；

(3-2) the diagnostic evidence vector E when k >=2, after renewal_1:k, its each element value is given by with following formula (1) and (2)

m_1:k(A)=α_km_1:k-1(A)+β_km_k(A|B)A,B∈Θ (1)

m_1:k(Θ)=1-∑_A∈Θm_1:k(A) (2)

Wherein, the m in formula (1)_1:k-1(A) represent the k-1 moment update after diagnostic evidence E_1:k-1Reliability assignment to fault A；m_k(A| B) represent the conditioning reliability assignment about fault A of kth moment acquisition, work as A=F_jTime, if diagnostic evidence E in k moment_k= (m_k(F₀),m_k(F₁),…,m_k(F_j),…,m_k(F_N),m_k(Θ) in), m_k(F_j) more than other all of m_k(F₀),m_k(F₁),…, m_k(F_j-1),…,m_k(F_j+1),…,m_k(F_N), then B=F_j, m_k(A | B)=1；Otherwise, m_k(A | B)=0；

α_kAnd β_kFor linear fusion smoothing weights, solution procedure is as follows:

A () be diagnostic evidence E after the acquisition k-1 moment updates_1:k-1, k moment diagnostic evidence E_kWith k+1 moment diagnostic evidence E_k+1It After, utilize formula (3) to calculate vector E_1:k-1With vector E_kBetween distance d (E_1:k-1,E_k) it is

$d(E_{1:k-1},E_k)=\sqrt{\frac{1}{2}(E_{1:k-1}-E_k)\underset{\‾}{\underset{\‾}{D}}{(E_{1:k-1}-E_k)}^T}---\left(3\right)$

E in formula_1:k-1-E_kBeing the vector obtained after two vector corresponding elements subtract each other, T represents the transposition of vector,It is one (N+2) The matrix of × (N+2), its diagonal entry value is 1, and the element value of N+1 row is walked in the 1st of its N+2 row, and 1st row of N+2 row are 1/ (N+1) to the element value of N+1 row, and other element values are 0；

In like manner, formula (4) vector E is obtained_1:k-1With vector E_k+1Between distance d (E_1:k-1,E_k+1) it is

$d(E_{1:k-1},E_{k+1})=\sqrt{\frac{1}{2}(E_{1:k-1}-E_{k+1})\underset{\‾}{\underset{\‾}{D}}{(E_{1:k-1}-E_{k+1})}^T}---\left(4\right)$

Vector E is obtained by formula (5)_kWith vector E_k+1Between distance d (E_k,E_k+1) it is

$d(E_k,E_{k+1})=\sqrt{\frac{1}{2}(E_k-E_{k+1})\underset{\‾}{\underset{\‾}{D}}{(E_k-E_{k+1})}^T}---\left(5\right)$

B () is tried to achieve distance d (E by above-mentioned steps (a)_1:k-1,E_k)、d(E_1:k-1,E_k+1) and d (E_k,E_k+1After), calculate the k moment E_1:k-1、E_kAnd E_k+1Similarity between any two:

Formula (6) is utilized to calculate vector E_1:k-1With vector E_kBetween similarity c (E_1:k-1,E_k) it is

c(E_1:k-1,E_k)=1-d (E_1:k-1,E_k) (6)

Similarity c (E_1:k-1,E_k) it is to weigh vector E_1:k-1With vector E_kClose degree, that is the degree that two evidences are consistent, And have c (E_1:k-1,E_k)=c (E_k,E_1:k-1), i.e. vector E_1:k-1With vector E_kSimilarity equal to vector E_kWith vector E_1:k-1Phase Like degree；

In like manner, formula (7) vector E is obtained_1:k-1With vector E_k+1Between similarity c (E_1:k-1,E_k+1) it is

c(E_1:k-1,E_k+1)=1-d (E_1:k-1,E_k+1) (7)

Vector E is obtained by formula (8)_kWith vector E_k+1Between similarity c (E_k,E_k+1) it is

c(E_k,E_k+1)=1-d (E_k,E_k+1) (8)

C () obtains at k moment diagnostic evidence vector E according to above-mentioned steps (b)_1:k-1、E_kAnd E_k+1Similarity c (E two-by-two_1:k-1, E_k), c (E_1:k-1,E_k+1) and c (E_k,E_k+1After), calculate the support that each evidence vector is supported by other two evidence vectors Degree:

Formula (9) is utilized to calculate evidence vector E_1:k-1By evidence vector E_kAnd E_k+1Support s (the E supported_1:k-1) it is

s(E_1:k-1)=c (E_1:k-1,E_k)+c(E_1:k-1,E_k+1) (9)

Support s is the function of similarity measurement, represents the degree that this evidence is supported by other evidences, s (E_1:k-1) value is the highest, Then explanation evidence vector E_1:k-1With evidence vector E_kAnd E_k+1Between similarity the highest；

In like manner, formula (10) evidence vector E is calculated_kBy evidence vector E_1:k-1And E_k+1Support s (the E supported_k) it is

s(E_k)=c (E_1:k-1,E_k)+c(E_k,E_k+1) (10)

Evidence vector E is calculated by formula (11)_k+1By evidence vector E_1:k-1And E_kSupport s (the E supported_k+1) it is

s(E_k+1)=c (E_1:k-1,E_k+1)+c(E_k,E_k+1) (11)

D () obtains diagnostic evidence vector E successively based on step (c)_1:k-1、E_kAnd E_k+1Reliability K in the k moment:

Formula (12) is utilized to calculate evidence vector E_1:k-1Reliability K (E in the k moment_1:k-1) it is

$K\left(E_{1:k-1}\right)=\frac{s\left(E_{1:k-1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(12\right)$

In like manner, formula (13) evidence vector E is calculated_kReliability K (E in the k moment_k) it is

$K\left(E_k\right)=\frac{s\left(E_k\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(13\right)$

Evidence vector E is calculated by formula (14)_k+1Reliability K (E in the k moment_k+1) it is

$K\left(E_{k+1}\right)=\frac{s\left(E_{k+1}\right)}{s\left(E_{1:k-1}\right)+s\left(E_k\right)+s\left(E_{k+1}\right)}---\left(14\right)$

There is K (E_1:k-1)+K(E_k)+K(E_k+1)=1, reliability K of an evidence vector is the highest, and this evidence and other evidences are described Similarity the highest, this evidence is the most reliable, and vice versa；

E E that () is tried to achieve based on above-mentioned steps (b)_1:k-1With E_k+1Between similarity c (E_1:k-1,E_k+1)、E_kWith E_k+1Between Similarity c (E_k,E_k+1), by judging that size therebetween determines α_k, β_kValue:

If c is (E_1:k-1,E_k+1)≥c(E_k,E_k+1),

If c is (E_1:k-1,E_k+1) ＜ c (E_k,E_k+1),

Try to achieve α_k, β_kValue after substituted in the formula (1) of step (3-2), each moment can be obtained more by recursive calculation Diagnostic evidence after Xin；

Step (4) is according to above-mentioned steps (3) diagnostic evidence E after the renewal that the k moment obtains_1:k=(m_1:k(F₀),m_1:k(F₁),…, m_1:k(F_j),…,m_1:k(F_N),m_1:k(Θ)), the diagnosing malfunction to rotating machinery: if m_1:k(F_j) value more than its He is m_1:k(F₀),m_1:k(F₁),…,m_1:k(F_j-1),…,m_1:k(F_j+1),…,m_1:k(F_N), then judge fault F_jOccur.