CN110097134B - Mechanical fault early diagnosis method based on time sequence - Google Patents

Mechanical fault early diagnosis method based on time sequence Download PDF

Info

Publication number
CN110097134B
CN110097134B CN201910383794.7A CN201910383794A CN110097134B CN 110097134 B CN110097134 B CN 110097134B CN 201910383794 A CN201910383794 A CN 201910383794A CN 110097134 B CN110097134 B CN 110097134B
Authority
CN
China
Prior art keywords
mechanical fault
mechanical
training sample
truncated
confidence
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
Application number
CN201910383794.7A
Other languages
Chinese (zh)
Other versions
CN110097134A (en
Inventor
吕俊伟
胡学钢
李培培
廖建兴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN201910383794.7A priority Critical patent/CN110097134B/en
Publication of CN110097134A publication Critical patent/CN110097134A/en
Application granted granted Critical
Publication of CN110097134B publication Critical patent/CN110097134B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a mechanical fault early diagnosis method based on a time sequence, which comprises the following steps: 1. cutting the characteristic data of the mechanical fault monitoring sequence data into s mechanical fault sequence data according to the length r; 2. selecting a mechanical fault classifier, training on each truncated training sample, and giving mechanical fault classification results of all mechanical fault characteristic data; 3. counting the accuracy of each type of mechanical fault classification on each truncated training sample, and calculating the trust of the mechanical fault classification result; 4. determining a confidence threshold theta of a mechanical fault classification result; 5. and carrying out early diagnosis on the mechanical fault test data by using the selected mechanical fault classifier and the confidence threshold theta. The method is suitable for the fault diagnosis of the mechanical fault sequence data with equal length and the mechanical fault sequence data with unequal length, and can pre-judge the type of the mechanical fault under the condition of ensuring the accuracy of the mechanical fault diagnosis.

Description

Mechanical fault early diagnosis method based on time sequence
Technical Field
The invention relates to the field of mechanical fault diagnosis, in particular to a mechanical fault early diagnosis method based on a time sequence.
Background
The mechanical failure diagnosis is a technology which can know and master the state of a machine in the operation process, determine the whole or local normality or abnormality of the machine, find a failure and the reason thereof at an early stage and forecast the development trend of the failure. Oil monitoring, vibration monitoring, noise monitoring, performance trend analysis, nondestructive inspection and the like are main diagnostic technical modes.
The signals for mechanical state monitoring and fault diagnosis mainly include vibration diagnosis, oil sample analysis, temperature monitoring and nondestructive testing flaw detection, and other technologies or methods are auxiliary. The vibration diagnosis is related to the widest field, the most advanced theoretical basis and the most sufficient research. In the aspect of analysis processing of vibration signals, besides classical statistical analysis, time-frequency domain analysis, time sequence model analysis and parameter identification, frequency refinement technology, cepstrum analysis, resonance demodulation analysis, three-dimensional holographic spectrum analysis, axis trajectory analysis, short-time fourier transform based on non-stationary signal hypothesis, Winger distribution, Hilbert-Huang transform, wavelet transform and the like have been developed recently. The research result of the current artificial intelligence injects new vitality into mechanical fault diagnosis, the expert system of fault diagnosis is quite developed theoretically, and has a successful application example, as an important branch of artificial intelligence, the research of artificial neural network becomes a latest research hotspot in the field of mechanical fault diagnosis.
The early diagnosis of the mechanical failure is a diagnosis and treatment method aiming at early failure of the machine and even possible failure, and aims to find early treatment so as to reduce the cost caused by the failure. Therefore, early diagnosis of mechanical failure is receiving increasing attention from the industry and academia.
The mechanical fault monitoring data is used as one of time sequence data, and a time sequence early classification method can be used for early diagnosis of mechanical faults. The current time series early classification methods can be roughly divided into two categories: rule-based methods and feature-based methods.
The rule-based method comprises the following steps: xing Zhengzheng et al propose that the 1NN classifier is used for early classification of time series, and give a concept of minimum prediction length, so that the classification accuracy can be ensured and the classification can be carried out in advance under the condition that the 1NN method is effective; recently, Uue Mori et al first obtain the probability of each sample at different time by using a probability classifier, and then propose a stopping rule based on the probability, and when the rule is satisfied, the classification result can be given in advance. However, current rule-based methods assume that the time series are of equal length, i.e. the length of the series and when it ends can be predicted in advance, which is not practical in the field of mechanical fault diagnosis.
The method based on the characteristics comprises the following steps: xing Zhengzheng et al proposed an early classification method based on interpretable feature maps, and the maps capable of distinguishing each class in advance were selected through two stages of feature extraction and feature selection. He guolang et al proposed an early classification method based on the multivariable time series of shape. However, at present, extraction and selection of shapeets are based on euclidean distance, in practical application, a mechanical fault sequence mostly has phase shift on a time axis, and the shapeets-based method is difficult to achieve high accuracy.
Therefore, at present, no time-series early classification method is well applicable to early diagnosis of mechanical faults, namely, on the premise of ensuring the accuracy of mechanical fault diagnosis, early diagnosis is performed as far as possible.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides the mechanical fault early diagnosis method based on the time sequence, is suitable for the mechanical fault sequence data with equal length and the fault diagnosis of the mechanical fault sequence data with unequal length, and achieves the aim of early diagnosis as far as possible on the premise of ensuring the accuracy of the mechanical fault diagnosis.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
the invention relates to a mechanical fault early diagnosis method based on time series, which is characterized by comprising the following steps of:
step 1: acquiring a set of mechanical fault sequence data as a training sample, wherein the training sample is formed by mechanical fault feature data TS ═ { TS ═ TS1,TS2,···,TSi,···,TSNY and mechanical failure classification label data Y ═ Y1,y2,···,yi,···,yNWhere N denotes the number of training samples, TSiRepresenting ith piece of mechanical fault characteristic data in the training sample, and comprising:
Figure GDA0002752959150000021
Lirepresenting the length, t, of the ith mechanical fault signature data in the training samplejRepresents the jth time, x, in the training samplei,jRepresenting a characteristic value corresponding to ith mechanical fault characteristic data in the training sample at jth time; y isiRepresenting ith mechanical fault characteristic data TS in the training sampleiThe corresponding mechanical fault classification label is provided with: y isi＝{cm1,2, C represents a mechanical fault classification standardNumber of labels, cmRepresents the m-th mechanical failure classification label, i belongs to [1, N ∈]，j∈[1,Li]；
Step 2: truncating the training samples into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s truncated training samples; wherein the d-th truncated training sample is
Figure GDA0002752959150000022
Figure GDA0002752959150000023
Representing the ith mechanical fault feature data in the d truncated training sample, and having
Figure GDA0002752959150000024
d∈[1,s]；
And step 3: selecting a classifier according to the characteristics of the training sample, and then carrying out TS (transport stream) on the d-th truncated training sampledInputting the training sample into the classifier, and obtaining a mechanical fault classifier f (TS) in a cross validation mode, thereby obtaining the d-th truncated training sample TS by using the mechanical fault classifier f (TS)dMechanical fault classification result of
Figure GDA0002752959150000025
Wherein
Figure GDA0002752959150000026
A classification result representing an ith mechanical fault feature data of the d truncated training sample;
and 4, step 4: calculating the mth mechanical failure classification label c of the d truncated training sample by using the formula (1)mMechanical fault classification accuracy Accd(cm)：
Figure GDA0002752959150000031
In the formula (1), | · | | represents the number of elements in the set,
Figure GDA0002752959150000032
the classification result of the ith mechanical fault feature data representing the d-th truncated training sample is equal to the m-th mechanical fault classification label cm，
Figure GDA0002752959150000033
The classification result of the ith mechanical fault feature data of the ith truncated training sample is equal to the actual classification label of the ith mechanical fault feature data of the kth truncated training sample;
and 5: determining a confidence threshold theta of the classification result:
step 5.1: calculating the Confidence factor of the classification result of the ith mechanical fault feature data in the d truncated training sample by using the formula (2)d(i) Thus obtaining the Confidence { Confidence of N × s classification resultsd(i)|d＝1，2，…，s；i＝1，2，…，N}：
Figure GDA0002752959150000034
In the formula (2), the reaction mixture is,
Figure GDA0002752959150000035
the h-th truncated training sample has a mechanical failure classification label of
Figure GDA0002752959150000036
The accuracy of the classification of the mechanical faults of the system,
Figure GDA0002752959150000037
a classification result representing an ith mechanical fault feature data of the h truncated training sample;
step 5.2: confidence { Confidence of Nx s classification resultsd(i) 1,2, …, s; i-1, 2, …, N, sorting in ascending order and removing duplicate values to obtain a set of processed confidence values
Figure GDA0002752959150000038
Wherein N iscNumber, v, representing confidence level after processingzRepresents the processed z-th confidence value, z is equal to [1, N ∈z]；
Step 5.3: calculating the midpoints of two adjacent elements in the processed confidence value Conf to obtain a group of candidate confidence threshold values
Figure GDA0002752959150000039
Wherein,
Figure GDA00027529591500000310
representing a z-th value in the candidate confidence threshold;
step 5.4: initializing z to 1, and enabling the minimum value f of the objective functionminMAX _ VALUE, where MAX _ VALUE is a constant;
step 5.5: if z is less than or equal to NcIf the number of training samples correctly classified is-1, the initialization stage executes step 5.6 after the initialization advance earliness is 0; otherwise, executing step 6;
step 5.6: initializing i to 1;
step 5.7: if i is not more than N, initializing d to 1, and executing the step 5.8; otherwise, the objective function f is calculated using equation (3):
f＝α×(N-correct)+(1-α)×earliness (3)
in the formula (3), alpha is a constant, and alpha is more than or equal to 0 and less than or equal to 1; if f < fminThen f is assigned to fmin，θzAssigning value to theta, z +1 to z, and executing step 5.5; if f ≧ fminIf so, assigning z +1 to z, and then executing step 5.5;
step 5.8: if the Confidence Confidence of the classification result of the ith mechanical fault feature data in the d truncated training sampled(i)＞θzOr d is more than s, calculating the updated advance earliness 'by using the formula (4) and assigning the updated advance earliness' to earliness, and then executing the step 5.9, otherwise, assigning d +1 to d, and then executing the step 5.8;
earliness′＝earliness+min(1.0,d×r/Li) (4)
in the formula (4), min is a minimum function;
step 5.9: if it is not
Figure GDA0002752959150000041
Assigning correct +1 to correct, assigning i +1 to i, then executing step 5.7, otherwise assigning i +1 to i, then executing step 5.7;
step 6: early diagnosis is carried out on the test sample of the mechanical fault sequence;
step 6.1: acquiring a new set of mechanical fault sequence data as a test sample, wherein the test sample is mechanical fault test characteristic data T ═ { T ═ T1,T2,···,Tk,···,TNtIn which N istRepresents the number of the test samples, TkRepresenting the characteristic data of the kth mechanical fault test in the test sample, and comprising:
Figure GDA0002752959150000045
Lkrepresenting the length, t, of the kth piece of mechanical failure test characteristic data in the test samplepRepresenting the p-th time in the test sample, ek,pRepresenting the characteristic value corresponding to the kth mechanical fault test characteristic data in the test sample at the p time, k is equal to [1, N ∈t]，p∈[1,Lk]；
Step 6.2: truncating the test sample into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s truncated test samples; wherein the d-th truncated test specimen
Figure GDA0002752959150000042
Figure GDA0002752959150000043
Representing characteristic data of a kth mechanical failure in said d-th truncated test sample, having
Figure GDA0002752959150000044
d∈[1,s]；
Step 6.3: using the mechanical fault classifier f (TS) at the d-th truncated test sample TdOn obtaining a classification result of the mechanical fault
Figure GDA0002752959150000051
Wherein
Figure GDA0002752959150000052
A mechanical fault classification result representing a kth piece of mechanical fault test feature data of the d-th truncated test sample;
step 6.4: calculating the Confidence Confidence of the classification result of the kth mechanical fault test characteristic data in the d truncated test sample by using the formula (2)d(k) If Confidenced(k) > theta, then
Figure GDA0002752959150000053
And (4) as a mechanical fault classification result of the kth mechanical fault test characteristic data in the test sample, otherwise, assigning d +1 to d, and returning to the step 6.3 for execution.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention provides a credibility concept of the mechanical fault classification result and provides a calculation formula of the credibility, thereby providing a quantitative measurement mode of the credibility of the classification result and facilitating the use of a user;
2. the invention provides a method for self-adapting the confidence threshold of the mechanical fault classification result, which does not need a user to set the threshold by himself and increases the usability of the method;
3. the invention does not limit the use of a specific mechanical fault sequence classifier, can select a proper classifier according to actual needs, and has weaker requirements on the mechanical fault sequence classifier compared with the mechanical fault sequence probability classifier.
4. The invention does not limit that the mechanical fault sequence is equal in length, not only can be suitable for the mechanical fault sequence with equal length, but also can be suitable for the mechanical fault sequence with unequal length, thereby ensuring that the method has wider applicability.
Drawings
FIG. 1 is a flow chart of the mechanical failure early diagnosis method based on time series according to the present invention;
FIG. 2 is a flow chart of the present invention for determining confidence thresholds for mechanical fault classification results.
Detailed Description
In this embodiment, referring to fig. 1, a method for early diagnosing a mechanical fault based on time series early classification is performed according to the following steps:
step 1: acquiring a group of mechanical fault sequence data as a training sample, wherein the training sample is formed by mechanical fault characteristic data TS ═ TS1,TS2,···,TSi,···,TSNY and mechanical failure classification label data Y ═ Y1,y2,···,yi,···,yNWhere N denotes the number of training samples, TSiRepresenting the ith piece of mechanical fault characteristic data in the training sample, and comprising:
Figure GDA0002752959150000054
Lirepresenting the length, t, of the ith mechanical fault signature data in the training samplejRepresenting the jth time, x, in the training samplei,jRepresenting a characteristic value corresponding to ith mechanical fault characteristic data in a training sample at jth time; y isiRepresenting ith mechanical fault characteristic data TS in training sampleiThe corresponding mechanical fault classification label is provided with: y isi＝{cm1,2, C represents the number of mechanical fault classification labels, CmRepresents the m-th mechanical failure classification label, i belongs to [1, N ∈]，j∈[1，Li](ii) a In specific implementation, N is 10, and 29 is less than or equal to Li361 percent or less, and C2 is 1 and 2 respectively.
Step 2: truncating the training samples into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s truncated training samples; wherein the d-th truncated training sample is
Figure GDA0002752959150000061
Figure GDA0002752959150000062
Representing the ith mechanical failure signature data in the d-th truncated training sample, and having
Figure GDA0002752959150000063
d∈[1，s](ii) a In a specific embodiment, r is 12 and s is 10. It should be noted that the ith mechanical failure feature data of the truncated training sample is also obtained from the 1 st time, but at the d × r times.
And step 3: after selecting a classifier according to the characteristics of the training samples, the d-th truncated training sample TS isdInputting the training sample into a classifier, and obtaining a mechanical fault classifier f (TS) in a cross validation mode, thereby obtaining a d-th truncated training sample TS by using the mechanical fault classifier f (TS)dMechanical fault classification result of
Figure GDA0002752959150000064
Wherein
Figure GDA0002752959150000065
A classification result representing ith mechanical failure feature data of the d truncated training sample;
in this embodiment, the classifier f (ts) selects the WEASEL algorithm, and the classification results of 10 feature data in the training sample are given in table 1.
TABLE 1 Classification results of 10 feature data in training samples
i d＝1 d＝2 d＝3 d＝4 d＝5 d＝6 d＝7 d＝8 d＝9 d＝10
1 1 2 1 2 2 2 2 2 2 2
2 2 1 2 1 1 1 1 1 1 1
3 2 2 1 1 1 1 1 1 1 1
4 2 2 2 1 1 1 1 1 1 1
5 1 1 1 1 1 1 1 1 1 1
6 1 2 2 2 2 2 2 2 2 2
7 1 2 2 2 2 2 2 2 2 2
8 2 2 2 2 2 2 2 2 2 2
9 2 2 2 2 2 2 2 2 2 2
10 2 1 1 2 2 2 2 2 2 2
And 4, step 4: computing the mth mechanical failure classification label c of the d-th truncated training sample by using equation (1)mMechanical fault classification accuracy Accd(cm)：
Figure GDA0002752959150000066
In the formula (1), | · | | represents the number of elements in the set,
Figure GDA0002752959150000067
the classification result of the ith mechanical fault feature data of the ith truncated training sample is an mth mechanical fault classification label cm，
Figure GDA0002752959150000068
The classification result of the ith mechanical failure feature data representing the d-th truncated training sample is equal to the actual classification result of the ith mechanical failure feature data of the d-th truncated training sampleClassifying labels; the meaning is as follows: denominator represents that the classification result on the d truncated training sample is the m-th mechanical fault classification label cmThe number of the (d) th truncated training sample is represented by the number of the (m) th mechanical fault classification label cmAnd the number of correct classification results;
in this embodiment, table 2 shows the mechanical fault classification accuracy of the mth mechanical fault classification label of the mth truncated training sample.
TABLE 2 mechanical Fault Classification accuracy of mth mechanical Fault Classification Label of the d-th truncated training sample
Figure GDA0002752959150000071
And 5: referring to fig. 2, a confidence threshold θ for the classification result is determined:
step 5.1: calculating the Confidence of the classification result of the ith mechanical fault feature data in the d truncated training sample by using the formula (2)d(i) Thus obtaining the Confidence { Confidence of N × s classification resultsd(i)|d＝1,2,…,s；i＝1,2,…,N}：
Figure GDA0002752959150000072
In the formula (2), the reaction mixture is,
Figure GDA0002752959150000073
the h-th truncated training sample has a mechanical failure classification label of
Figure GDA0002752959150000074
The accuracy of the classification of the mechanical faults of the system,
Figure GDA0002752959150000075
representing the classification result of the ith mechanical fault feature data of the h truncated training sample;
in this embodiment, take i ═ 1 as an example, Confidence1(1)＝0.25，Confidence2(1)＝0.71，Confidence3(1)＝1-(1-0.25)×(1-0.25)＝0.4375，Confidence4(1)＝1-(1-0.71)×(1-1)＝1，…
Step 5.2: confidence { Confidence of Nx s classification resultsd(i) 1,2, …, s; i-1, 2, …, N, sorting in ascending order and removing duplicate values to obtain a set of processed confidence values
Figure GDA0002752959150000076
Wherein N iscNumber, v, representing confidence level after processingzRepresents the processed z-th confidence value, z is equal to [1, N ∈z]；
Step 5.3: calculating the midpoints of two adjacent elements in the processed confidence value Conf to obtain a group of candidate confidence threshold values
Figure GDA0002752959150000077
Wherein,
Figure GDA0002752959150000078
representing a z-th value in the candidate confidence threshold;
step 5.4: initializing z to 1, and enabling the minimum value f of the objective functionminMAX _ VALUE, where MAX _ VALUE is a constant; in this embodiment, MAX _ VALUE is 9999999.
Step 5.5: if z is less than or equal to NcIf the number of training samples correctly classified is-1, the initialization stage executes step 5.6 after the initialization advance earliness is 0; otherwise, executing step 6;
step 5.6: initializing i to 1;
step 5.7: if i is not more than N, initializing d to 1, and executing the step 5.8; otherwise, the objective function f is calculated using equation (3):
f＝α×(N-correct)+(1-α)×earliness (3)
in the formula (3), alpha is a constant, and alpha is more than or equal to 0 and less than or equal to 1; if f < fminThen f is assigned to fmin，θzAssign theta, z +1 to z, and perform the steps5.5; if f ≧ fminAssigning z +1 to z, and then executing step 5.5; in this example, α is 0.8.
Step 5.8: if the Confidence Confidence of the classification result of the ith mechanical fault feature data in the d truncated training sampled(i)＞θzOr d is more than s, calculating the updated advance earliness 'by using the formula (4) and assigning the updated advance earliness' to earliness, and then executing the step 5.9, otherwise, assigning d +1 to d, and then executing the step 5.8;
earliness′＝earliness+min(1.0,d×r/Li) (4)
in the formula (4), min is a minimum function;
step 5.9: if it is not
Figure GDA0002752959150000081
Assigning correct +1 to correct, assigning i +1 to i, then executing step 5.7, otherwise assigning i +1 to i, then executing step 5.7; in this embodiment, the finally determined confidence threshold θ is 0.9.
Step 6: early diagnosis is carried out on the test sample of the mechanical fault sequence;
step 6.1: acquiring a group of new mechanical fault sequence data as a test sample, wherein the test sample is mechanical fault test characteristic data
Figure GDA0002752959150000082
Wherein N istIndicates the number of test specimens, TkThe characteristic data of the k-th mechanical failure test in the test sample are shown, and the following are provided:
Figure GDA0002752959150000083
Lkrepresenting the length, t, of the kth piece of mechanical failure test signature data in the test samplepDenotes the p-th time in the test sample, ek,pRepresenting the characteristic value corresponding to the kth mechanical fault test characteristic data in the test sample at the p time, k is equal to [1, N ∈t]，p∈[1,Lk](ii) a In this example, where Nt＝10，30≤Lk≤321。
Step 6.2: cutting the test sample into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s cut test samples; wherein the d-th truncated test specimen
Figure GDA0002752959150000084
Figure GDA0002752959150000085
Representing characteristic data of the kth mechanical failure in the d-th truncated test sample, having
Figure GDA0002752959150000091
d∈[1，s]；
Step 6.3: using mechanical fault classifier f (TS) at d truncated test sample TdOn obtaining a classification result of the mechanical fault
Figure GDA0002752959150000092
Wherein
Figure GDA0002752959150000093
A mechanical fault classification result representing the kth mechanical fault test characteristic data of the d-th truncated test sample;
in this embodiment, taking k as 1 as an example, the mechanical failure classification results of the k as 1-th mechanical failure test feature data of the d-th truncated test sample are shown in table 3.
TABLE 3 mechanical Fault Classification results of 1 st mechanical Fault test characteristic data of the d-th truncated test sample
k d＝1 d＝2 d＝3 d＝4 d＝5 d＝6 d＝7 d＝8 d＝9 d＝10
1 1 1 2 2 2 2 2 2 2 2
Step 6.4: calculating the Confidence of the classification result of the kth mechanical fault test characteristic data in the d-th truncated test sample by using the formula (2)d(k) If Confidenced(k) > theta, then
Figure GDA0002752959150000094
And (3) as a mechanical fault classification result of the kth mechanical fault test characteristic data in the test sample, otherwise, assigning d +1 to d, and returning to the stepStep 6.3 is performed.
In this embodiment, Confidence1(1)＝0.25＜θ，Confidence2(1)＝1-(1-0.25)×(1-0.67)＝0.7525＜θ，Confidence3(1)＝0.67＜θ，Confidence4(1) The mechanical failure classification result when d is 4 because 1- (1-0.67) × (1-1) > θ
Figure GDA0002752959150000095
And (3) as a mechanical fault classification result of 1 st mechanical fault test characteristic data.

Claims (1)

1. A mechanical fault early diagnosis method based on time series is characterized by comprising the following steps:
step 1: acquiring a set of mechanical fault sequence data as a training sample, wherein the training sample is formed by mechanical fault feature data TS ═ { TS ═ TS1,TS2,···,TSi,···,TSNY and mechanical failure classification label data Y ═ Y1,y2,···,yi,···,yNWhere N denotes the number of training samples, TSiRepresenting ith piece of mechanical fault characteristic data in the training sample, and comprising: TS (transport stream)i＝{(t1,xi,1),(t2,xi,2),···,(tj,xi,j),···,(tLi,xi,Li)}，LiRepresenting the length, t, of the ith mechanical fault signature data in the training samplejRepresents the jth time, x, in the training samplei,jRepresenting a characteristic value corresponding to ith mechanical fault characteristic data in the training sample at jth time; y isiRepresenting ith mechanical fault characteristic data TS in the training sampleiThe corresponding mechanical fault classification label is provided with: y isi＝{cm1,2, C represents the number of mechanical fault classification labels, CmRepresents the m-th mechanical failure classification label, i belongs to [1, N ∈]，j∈[1,Li]；
Step 2: truncating the training sample into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s truncationsBroken training samples; wherein the d-th truncated training sample is
Figure FDA0002834340810000011
Figure FDA0002834340810000012
Representing the ith mechanical fault feature data in the d truncated training sample, and having
Figure FDA0002834340810000013
And step 3: selecting a classifier according to the characteristics of the training sample, and then carrying out TS (transport stream) on the d-th truncated training sampledInputting the training sample into the classifier, and obtaining a mechanical fault classifier f (TS) in a cross validation mode, thereby obtaining the d-th truncated training sample TS by using the mechanical fault classifier f (TS)dMechanical fault classification result of
Figure FDA0002834340810000014
Wherein
Figure FDA0002834340810000015
A classification result representing an ith mechanical fault feature data of the d truncated training sample;
and 4, step 4: calculating the mth mechanical failure classification label c of the d truncated training sample by using the formula (1)mMechanical fault classification accuracy Accd(cm)：
Figure FDA0002834340810000016
In the formula (1), | · | | represents the number of elements in the set,
Figure FDA0002834340810000017
score of ith mechanical fault signature data representing the d-th truncated training sampleClass result is equal to the mth mechanical failure classification label cm，
Figure FDA0002834340810000018
The classification result of the ith mechanical fault feature data of the ith truncated training sample is equal to the actual classification label of the ith mechanical fault feature data of the kth truncated training sample;
and 5: determining a confidence threshold theta of the classification result:
step 5.1: calculating the Confidence factor of the classification result of the ith mechanical fault feature data in the d truncated training sample by using the formula (2)d(i) Thus obtaining the Confidence { Confidence of N × s classification resultsd(i)|d＝1,2,…,s；i＝1,2,…,N}：
Figure FDA0002834340810000021
In the formula (2), the reaction mixture is,
Figure FDA0002834340810000022
the mechanical failure classification label representing the h-th truncated training sample is
Figure FDA0002834340810000023
The accuracy of the classification of the mechanical faults of the system,
Figure FDA0002834340810000024
representing the classification result of the ith mechanical fault feature data of the h truncated training sample;
step 5.2: confidence { Confidence of Nx s classification resultsd(i) 1,2, …, s; i-1, 2, …, N, sorting in ascending order and removing duplicate values to obtain a set of processed confidence values
Figure FDA0002834340810000025
Wherein N iscAfter the representation treatmentNumber of degrees of trust, vzRepresents the processed z-th confidence value, z is equal to [1, N ∈z]；
Step 5.3: calculating the midpoints of two adjacent elements in the processed confidence value Conf to obtain a group of candidate confidence threshold values
Figure FDA0002834340810000026
Wherein,
Figure FDA0002834340810000027
representing a z-th value in the candidate confidence threshold;
step 5.4: initializing z to 1, and enabling the minimum value f of the objective functionminMAX _ VALUE, where MAX _ VALUE is a constant;
step 5.5: if z is less than or equal to NcIf the number of training samples correctly classified is-1, the initialization stage executes step 5.6 after the initialization advance earliness is 0; otherwise, executing step 6;
step 5.6: initializing i to 1;
step 5.7: if i is not more than N, initializing d to 1, and executing the step 5.8; otherwise, the objective function f is calculated using equation (3):
f＝α×(N-correct)+(1-α)×earliness (3)
in the formula (3), alpha is a constant, and alpha is more than or equal to 0 and less than or equal to 1; if f < fminThen f is assigned to fmin，θzAssigning value to theta, z +1 to z, and executing step 5.5; if f ≧ fminIf so, assigning z +1 to z, and then executing step 5.5;
step 5.8: if the Confidence Confidence of the classification result of the ith mechanical fault feature data in the d truncated training sampled(i)＞θzOr d is more than s, calculating the updated advance earliness 'by using the formula (4) and assigning the updated advance earliness' to earliness, and then executing the step 5.9, otherwise, assigning d +1 to d, and then executing the step 5.8;
earliness′＝earliness+min(1.0,d×r/Li) (4)
in the formula (4), min is a minimum function;
step 5.9: if it is not
Figure FDA0002834340810000031
Assigning correct +1 to correct, assigning i +1 to i, then executing step 5.7, otherwise assigning i +1 to i, then executing step 5.7;
step 6: early diagnosis is carried out on the test sample of the mechanical fault sequence;
step 6.1: acquiring a group of new mechanical fault sequence data as a test sample, wherein the test sample is mechanical fault test characteristic data
Figure FDA0002834340810000032
Wherein N istRepresents the number of the test samples, TkRepresenting the characteristic data of the kth mechanical fault test in the test sample, and comprising:
Figure FDA0002834340810000033
Lkrepresenting the length, t, of the kth piece of mechanical failure test characteristic data in the test samplepRepresenting the p-th time in the test sample, ek,pRepresenting the characteristic value corresponding to the kth mechanical fault test characteristic data in the test sample at the p time, k is equal to [1, N ∈t]，p∈[1,Lk]；
Step 6.2: truncating the test sample into s pieces of mechanical fault sequence data according to the length r, thereby obtaining s truncated test samples; wherein the d-th truncated test specimen
Figure FDA0002834340810000034
Figure FDA0002834340810000035
Representing characteristic data of a kth mechanical failure in said d-th truncated test sample, having
Figure FDA0002834340810000036
Figure FDA0002834340810000037
Step 6.3: using the mechanical fault classifier f (TS) at the d-th truncated test sample TdOn obtaining a classification result of the mechanical fault
Figure FDA0002834340810000038
Wherein
Figure FDA0002834340810000039
A mechanical fault classification result representing a kth piece of mechanical fault test feature data of the d-th truncated test sample;
step 6.4: calculating the Confidence Confidence of the classification result of the kth mechanical fault test characteristic data in the d truncated test sample by using the formula (2)d(k) If Confidenced(k) > theta, then
Figure FDA00028343408100000310
And (4) as a mechanical fault classification result of the kth mechanical fault test characteristic data in the test sample, otherwise, assigning d +1 to d, and returning to the step 6.3 for execution.
CN201910383794.7A 2019-05-08 2019-05-08 Mechanical fault early diagnosis method based on time sequence Active CN110097134B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910383794.7A CN110097134B (en) 2019-05-08 2019-05-08 Mechanical fault early diagnosis method based on time sequence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910383794.7A CN110097134B (en) 2019-05-08 2019-05-08 Mechanical fault early diagnosis method based on time sequence

Publications (2)

Publication Number Publication Date
CN110097134A CN110097134A (en) 2019-08-06
CN110097134B true CN110097134B (en) 2021-03-09

Family

ID=67447416

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910383794.7A Active CN110097134B (en) 2019-05-08 2019-05-08 Mechanical fault early diagnosis method based on time sequence

Country Status (1)

Country Link
CN (1) CN110097134B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111913463B (en) * 2020-06-12 2021-07-13 西安交通大学 State monitoring method for chemical volume control system of nuclear power plant
CN113608146B (en) * 2021-08-06 2023-12-19 云南电网有限责任公司昆明供电局 Fault line selection method suitable for forest fire under high-resistance grounding condition

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101788378A (en) * 2009-01-23 2010-07-28 西门子公司 Mechanical failure diagnostic method and device
CN108681747A (en) * 2018-05-11 2018-10-19 武汉理工大学 Rotary machinery fault diagnosis based on deep learning and condition monitoring system and method
CN108827605A (en) * 2018-03-20 2018-11-16 南京航空航天大学 A kind of mechanical breakdown characteristic automatic extraction method based on improvement sparseness filtering
CN109033632A (en) * 2018-07-26 2018-12-18 北京航空航天大学 A kind of trend forecasting method based on depth quantum nerve network
CN109685376A (en) * 2018-12-26 2019-04-26 国家电网公司华中分部 A kind of power customer abnormal behaviour method for early warning based on similarity analysis theory

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104091070B (en) * 2014-07-07 2017-05-17 北京泰乐德信息技术有限公司 Rail transit fault diagnosis method and system based on time series analysis
CN106371422B (en) * 2016-08-31 2018-08-31 北京航空航天大学 A method of prediction critical infrastructures fault propagation
CN107341504A (en) * 2017-06-07 2017-11-10 同济大学 A kind of Trouble Diagnostic Method of Machinery Equipment based on the popular study of time series data
CN108520080B (en) * 2018-05-11 2020-05-05 武汉理工大学 Ship diesel generator fault prediction and health state online evaluation system and method
CN108960077A (en) * 2018-06-12 2018-12-07 南京航空航天大学 A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network
CN109035488A (en) * 2018-08-07 2018-12-18 哈尔滨工业大学（威海） Aero-engine time series method for detecting abnormality based on CNN feature extraction
CN109308519A (en) * 2018-09-29 2019-02-05 广州博通信息技术有限公司 A kind of refrigeration equipment failure prediction method neural network based
CN109376799A (en) * 2018-11-29 2019-02-22 中国地质大学（武汉） A kind of time series early stage classification method and equipment based on diversity shapelet

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101788378A (en) * 2009-01-23 2010-07-28 西门子公司 Mechanical failure diagnostic method and device
CN108827605A (en) * 2018-03-20 2018-11-16 南京航空航天大学 A kind of mechanical breakdown characteristic automatic extraction method based on improvement sparseness filtering
CN108681747A (en) * 2018-05-11 2018-10-19 武汉理工大学 Rotary machinery fault diagnosis based on deep learning and condition monitoring system and method
CN109033632A (en) * 2018-07-26 2018-12-18 北京航空航天大学 A kind of trend forecasting method based on depth quantum nerve network
CN109685376A (en) * 2018-12-26 2019-04-26 国家电网公司华中分部 A kind of power customer abnormal behaviour method for early warning based on similarity analysis theory

Also Published As

Publication number Publication date
CN110097134A (en) 2019-08-06

Similar Documents

Publication Publication Date Title
CN112036301B (en) Driving motor fault diagnosis model construction method based on intra-class feature transfer learning and multi-source information fusion
CN105738109B (en) Bearing fault classification diagnosis method based on rarefaction representation and integrated study
Yu et al. A bearing fault diagnosis technique based on singular values of EEMD spatial condition matrix and Gath-Geva clustering
CN104899327B (en) A kind of time series method for detecting abnormality of no class label
CN109186973A (en) A kind of mechanical failure diagnostic method of unsupervised deep learning network
CN112729835B (en) Multi-domain semi-supervised fault diagnosis method and device for axial plunger pump bearing
CN110097134B (en) Mechanical fault early diagnosis method based on time sequence
Mechefske et al. Fault detection and diagnosis in low speed rolling element bearings Part II: The use of nearest neighbour classification
CN110647830A (en) Bearing fault diagnosis method based on convolutional neural network and Gaussian mixture model
CN111964909A (en) Rolling bearing operation state detection method, fault diagnosis method and system
CN112926642B (en) Multi-source domain self-adaptive intelligent mechanical fault diagnosis method and system
Kwon et al. Multi-scale speaker embedding-based graph attention networks for speaker diarisation
CN108470194B (en) Feature screening method and device
CN116595317A (en) Motor fault diagnosis method based on orthogonal stator current GA analysis and SFO-PNN
CN113252322A (en) Method and device for diagnosing mechanical fault of circuit breaker under noise-containing background
CN110987866A (en) Gasoline property evaluation method and device
CN116012681A (en) Method and system for diagnosing motor faults of pipeline robot based on sound vibration signal fusion
Huo et al. A class-level matching unsupervised transfer learning network for rolling bearing fault diagnosis under various working conditions
Zhao et al. Hybrid semi-supervised learning for rotating machinery fault diagnosis based on grouped pseudo labeling and consistency regularization
CN117746084A (en) Unsupervised domain adaptive pedestrian re-identification method based on attention residual error and contrast learning
CN110222390B (en) Gear crack identification method based on wavelet neural network
CN117312980A (en) Rotary equipment fault diagnosis method based on partial domain adaptation and knowledge distillation
CN116364108A (en) Transformer voiceprint detection method and device, electronic equipment and storage medium
CN114383846B (en) Bearing composite fault diagnosis method based on fault label information vector
CN111474476A (en) Motor fault prediction method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant