CN110110768A - Fault Diagnosis of Roller Bearings based on Concurrent Feature study and multi-categorizer - Google Patents
Fault Diagnosis of Roller Bearings based on Concurrent Feature study and multi-categorizer Download PDFInfo
- Publication number
- CN110110768A CN110110768A CN201910331462.4A CN201910331462A CN110110768A CN 110110768 A CN110110768 A CN 110110768A CN 201910331462 A CN201910331462 A CN 201910331462A CN 110110768 A CN110110768 A CN 110110768A
- Authority
- CN
- China
- Prior art keywords
- model
- training sample
- encoding encoder
- label
- stacking
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- 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)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention proposes a kind of rolling bearing intelligent failure diagnosis methods learnt based on Multi-classifers integrated and Concurrent Feature, it is intended to improve the nicety of grading of model, realize step are as follows: obtain training sample set and test sample collection;Multiple storehouse self-encoding encoder models are established, is that input carries out parallel training to stacking-type self-encoding encoder model with training sample set, extracts multiple features of training sample set;Characteristic evaluating is carried out based on feature of the softmax model to extraction, feature constitutive characteristic subset is screened according to corresponding threshold value and evaluation index value;Multiple classifiers based on softmax model are established according to character subset, it is the nicety of grading that input obtains each classifier with character subset, multiple classifiers, which are reselected, according to threshold value constructs integrated multi-categorizer model, integrated multi-categorizer model prediction label is obtained by majority voting method, prediction label and rolling bearing fault type are mapped, realize the intelligent trouble diagnosis of rolling bearing.
Description
Technical field
The invention belongs to rotating machinery intelligent Fault Diagnosis Technique fields, are related to a kind of Fault Diagnosis of Roller Bearings,
More particularly to a kind of rolling bearing fault intelligent diagnosing method based on Concurrent Feature study and integrated multi-categorizer, can be used for rolling
The automatic fault diagnosis of the rotating machineries such as dynamic bearing.
Background technique
Rotating machinery plays an important role in industrial equipment.Rolling bearing is motor, wind power generating set and gear-box
One of most important component in equal rotating machineries, it is made of rolling element, outer ring, inner ring and retainer.Rolling bearing usually exists
It works under complicated operating condition, such as different operating conditions, vibration, temperature, load, these factors frequently can lead to rolling bearing
The decline of performance even malfunction and failure.The performance state of rolling bearing directly affects the operational safety of equipment, therefore, automatic,
The malfunction for accurately diagnosing rolling bearing is extremely important.
Rolling bearing fault diagnosis be mainly by being run to rolling bearing when some dynamic parameters for example temperature, amplitude,
The signals such as displacement are analyzed and processed, and are identified to the data of rolling bearing difference operating condition, to reach the mesh of fault diagnosis
's.In general, the index for evaluating a kind of Fault Diagnosis of Roller Bearings quality has diagnostic accuracy, diagnosis efficiency, robustness, objective
Property etc..Fault Diagnosis of Roller Bearings can be divided into conventional fault diagnosis method and intelligent failure diagnosis method.Intelligent diagnostics
Method generally comprises three steps: 1) data acquire, and 2) feature extraction and selection, 3) failure modes diagnosis.According to feature extraction
Different with the method that selection course uses, intelligent diagnosing method can be divided into the intelligent failure diagnosis method based on shallow-layer feature learning
With the intelligent failure diagnosis method learnt based on further feature.
Conventional fault diagnosis method is mostly based on signal processing technology using physical model and establishes fault diagnosis model, such as experience
Mode decomposition and wavelet decomposition.However, in practical engineering applications original vibration signal often show it is complicated, non-linear and more
The characteristics of noise, needs to rely at advanced signal for the Accurate Diagnosis of fault type, fault severity level and fault direction
Reason technology, in addition, the accurate description needs of the bearing performance state under complex working condition extract a large amount of time domains, frequency from original signal
Domain and time and frequency domain characteristics, usually from these features selection with diagnosis target correlation by force, more representative feature be one
Blindly, subjective and time-consuming work lacks visitor so traditional method for diagnosing faults carries out feature selecting dependent on expertise
The property seen, it is difficult to which the bearing fault state under complex working condition in Practical Project automatically, accurately identify.
Rolling bearing intelligent failure diagnosis method is to be got up based on data-driven with sensor and technical development of computer
A kind of method, as support vector machines, principal component analysis, artificial neural network, stacking-type self-encoding encoder, convolutional neural networks,
Depth confidence network etc..Wherein, although the intelligent diagnosing methods such as support vector machines, principal component analysis and artificial neural network can be put
The de- dependence to expertise, realizes the adaptive learning of rolling bearing performance state feature, improves fault diagnosis result
Objectivity, still, this method for diagnosing faults is a kind of intelligent failure diagnosis method based on shallow-layer feature learning, be difficult from
Further feature is extracted in initial data.Therefore, the feature learning ability of this method is weak, and failure modes diagnostic accuracy is low.
In order to improve the feature learning ability of model, scholars propose with stacking-type self-encoding encoder, convolutional neural networks,
Depth confidence network etc. is the intelligent failure diagnosis method based on further feature study of representative.Wherein, due to network structure
The training process of complexity, convolutional neural networks and depth confidence network is extremely complex, and stacking-type autocoder is due to its knot
Structure simply with unsupervised ability in feature extraction, is widely used in the feature extraction in the fields such as pattern-recognition, fault diagnosis
Cheng Zhong.In addition, softmax classifier is often used as intelligence and examines when carrying out failure modes diagnosis using intelligent diagnostics model
The last layer of disconnected model realizes feature the reflecting to label extracted to stacking-type autocoder to obtain the output of model
It penetrates.For example, Shao Haidong et al. is in 2018 on volume 102 of Mechanical Systems and Signal Processing
" the A novel method for intelligent fault diagnosis of rolling bearings delivered
In the article of using ensemble deep auto-encoders ", a kind of rolling of integrated depth self-encoding encoder model is proposed
Dynamic bearing intelligent failure diagnosis method, this method acquire the vibration data of rolling bearing first, and divide training set and test set,
Secondly, establishing integrated depth self-encoding encoder model based on different activation primitives and being instructed in advance using training set data to model
Practice, network is finely adjusted using faulty tag on this basis, finally, exporting test specimens using a softmax classifier
This prediction label, realize according to the bearing vibration time-domain signal of field real-time acquisition to the malfunction of rolling bearing into
Row diagnosis, provides reference for the safe operation and maintenance of rotating machinery.However, although this method is in feature extraction and selection
Stage can extract further feature by integrated depth self-encoding encoder model, but not carry out to the further feature extracted effective
Evaluation cannot filter out and diagnose that target correlation is strong, more representative further feature, affect the classification diagnosis essence of model
Degree;In addition, this method is diagnosed in failure modes diagnostic phases merely with a softmax classifier, poor robustness, with
Cause the diagnosis accuracy of model low under machine interference.
Summary of the invention
It is an object of the invention to overcome the problems of the above-mentioned prior art, provide a kind of based on Concurrent Feature study
With the rolling bearing intelligent failure diagnosis method of integrated multi-categorizer, it is intended to improve the fault diagnosis precision of rolling bearing.
Technical thought of the invention is to acquire bearing vibration acceleration time domain signal first, obtains training sample set
And test sample collection;Secondly, establishing multiple stacking-type self-encoding encoder models based on different activation primitives, made with training sample set
For the input of stacking-type self-encoding encoder model, each stacking-type self-encoding encoder model is trained, extracts training sample parallel
Multiple features of collection;Then, characteristic evaluating is carried out to each feature extracted parallel based on Softmax model, according to corresponding
The Performance Evaluating Indexes value of threshold value and each feature establishes base further according to character subset using the feature filtered out as character subset
In multiple classifiers of Softmax model, it is input with character subset, the nicety of grading of each classifier is obtained, according to classification
Device precision threshold, reselects multiple classifiers and constructs integrated multi-categorizer model, integrates more points by majority voting method acquisition
The output label of class device model, to establish rolling bearing fault diagnosis model;Test sample collection is finally inputted into rolling bearing
In fault diagnosis model, the prediction label of test sample is obtained, prediction label is mapped back to the fault type of rolling bearing, is realized
To the fault diagnosis of rolling bearing.
To achieve the goals above, the technical solution adopted by the present invention includes the following steps:
(1) training sample set and test sample collection are obtained:
(1a) using I vibration time-domain signal data of the rolling bearing chosen from database as training sample, each
Training sample includes a kind of label for indicating fault category, and the classification sum of label is Q, all training sample composing training samples
Collect X1,Wherein, I >=2000, and I > > Q, xiIndicate i-th of training sample, y(i)Indicate xiLabel;
(1b) makees J vibration time-domain signal data of the rolling bearing to be diagnosed acquired in real time by data collection system
For test sample, all test samples constitute test sample collection X2,J≥I/2,xjIndicate j-th of test sample;
(2) N number of stacking-type self-encoding encoder model is constructed:
Constructed based on different activation primitive othernesses it is N number of respectively include K self-encoding encoder stacking-type self-encoding encoder model,
K-th of self-encoding encoder is denoted as in n-th of stacking-type self-encoding encoder model Hidden layer beN number of stacking-type is self-editing
The number of nodes of the last one hidden layer of code device model is h, and the number of nodes of output layer is o, wherein n=1,2 ..., N, k
=1,2 ..., K, N >=2, K >=2;
(3) parallel training is carried out to N number of stacking-type self-encoding encoder model:
(3a) enables n=1, k=1;
(3b) is by training sample set X1As k-th of self-encoding encoder in n-th of stacking-type self-encoding encoder modelInput,
It is rightIt is trained, the hidden layer after being trained isSelf-encoding encoder
(3c) is by the self-encoding encoder after trainingHidden layerAs kth+1 in n-th of stacking-type self-encoding encoder model
A self-encoding encoderInput, it is rightIt is trained, the hidden layer after being trained isSelf-encoding encoder
Whether (3d) judges k=K true, if so, otherwise n-th of stacking-type self-encoding encoder model after being trained enables
K=k+1, and execute step (3c)~(3d);
Whether (3e) judges n=N true, if so, the stacking-type self-encoding encoder model after N number of parallel training is obtained, otherwise,
N=n+1 is enabled, and executes step (3b)~(3d);
(4) multiple performance state features are obtained:
It will be in the stacking-type self-encoding encoder model after n-th of parallel trainingH node output valve be used as from training sample
The n-th group performance state feature that this concentration is extracted, the performance state feature that all stacking-type self-encoding encoders are extracted is M altogether
It is a, wherein M=N × h, m-th of performance state feature are Fm, m=1,2 ..., M;
(5) to each performance state feature FmCarry out characteristic evaluating:
Softmax disaggregated model is trained by minimizing error function, the softmax classification mould after being trained
Type, and by each performance state feature FmSoftmax disaggregated model after being input to training, obtains M performance evaluation index value
I1,I2,…,Im,…,IM;
(6) the integrated multi-categorizer model based on softmax classifier is established:
Q threshold value T is arranged in (6a)1,T2,…,Tq,…,TQ, and by M performance evaluation index value one by one with q-th of threshold value Tq
It is compared, filters out Performance Evaluating Indexes value greater than TqPerformance state feature composition characteristic subset Sq, obtain Q feature
Collect S1,S2,…,Sq,…,SQ, q=1,2 ..., Q;
(6b) is by q-th of character subset SqIt is input to softmax classifier CqIn, obtain CqPrediction label vector Rq,And calculate CqClassification diagnosis precision Acq, obtain the classification diagnosis precision Ac of Q softmax classifier1,
Ac2,…,Acq,…,AcQ;
Threshold value T is arranged in (6c)c, by Q classification diagnosis precision Ac1,Ac2,…,Acq,…,AcQRespectively with TcCompare, screens
Classification diagnosis precision is greater than T outcλ classifier C1,C2,…,Cκ,…,Cλ, λ >=5 constitute the collection based on softmax classifier
At multi-categorizer MODEL C;
(7) rolling bearing fault diagnosis result is obtained:
(7a) is by X2In xjIt is input to each of C Cκ, κ=1,2 ..., λ obtain λ classifier C1,C2,…,
Cκ,…,CλTo xjPrediction labelWherein,
(7b) searches prediction label by ballot methodThe most prediction label of middle frequency of occurrence, and
As the integrated multi-categorizer MODEL C based on softmax classifier to test sample xjPrediction label L(j):
Wherein, mode () intermediate scheme searches function,It indicates using the κ classifier filtered out to test sample
xjThe label of prediction, λ indicate the number of the classifier filtered out;
Step 7c) by prediction label L(j)The fault category for including with training sample set is mapped, and is obtained rolling bearing and is existed
The malfunction of different moments.
Compared with the prior art, the invention has the following advantages:
The present invention concurrently extracts training sample from encoding model in feature extraction and choice phase, using multiple stacking-types
Multiple performance state features, and it is strong, more representative by the setting of characteristic evaluating and threshold value to filter out target correlation
The character subset filtered out is input to multiple Softmax and classified by feature as character subset, and in failure modes diagnostic phases
In device, the higher multiple classifiers of diagnostic accuracy are reselected according to the diagnostic accuracy of Softmax classifier and classifier threshold value,
Avoid the prior art is had compared with prior art using the defect that single classifier carries out failure modes diagnosis poor robustness
Improve to effect the accuracy of the fault diagnosis under random disturbances.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention;
Fig. 2 is the vibration time domain signal waveform schematic diagram of 12 kinds of different faults types of rolling bearing of the present invention;
Fig. 3 is the character subset schematic diagram that the present invention screens;
Fig. 4 is the rolling bearing intelligent trouble diagnosis result schematic diagram that the present invention is implemented.
Specific embodiment
In the following with reference to the drawings and specific embodiments, the present invention is described in further detail:
Referring to Fig.1, the present invention includes the following steps:
Step 1) obtains training sample set and test sample collection
Rolling bearing, which is acquired, by data collection system amounts to 12 kinds of fault types, 3600 vibration time-domain signals as number
According to collection, and will wherein for 2400 data as training set, remaining 1200 data is specific as follows as test set:
The vibration time-domain signal that the present embodiment uses is all from the bearing of bearing accelerated life test platform PRONOSTIA acquisition
Vibrate time-domain signal.The platform consists of three parts: drive module, and load blocks sum number adopts module.The experimental rig it is main
Function is to provide the signal of different faults type, and the main component of experimental provision includes a driving motor, a torque sensing
Device and a dynamometer machine, driving motor power 1.2Kw, maximum (top) speed 6000r/min.Bearing designation is 6205-2RS JEM
SKF, acceleration transducer (DYTRAN 3035B) are mounted near driving end, sample frequency 12kHz.Working condition are as follows: turn
Fast 1800rpm, load 4000N.Test bearing mainly includes that normal condition, segment sunken (BD), outer ring defect (OR) and inner ring lack
Fall into (IR) four kinds of fault types.Single Point of Faliure is introduced into test bearing using electrical discharge machining, fault diameter includes 0.007,
0.014,0.021 and 0.028 inch, totally four kinds of Dimension Types, obtain including different fault types, different faults diameter ruler
The bearing vibration time-domain signal of very little and different faults orientation total 12 kinds of fault types, waveform are as shown in Figure 2.For
Every kind of fault type, generates 300 samples from original vibration signal, and data point is 400.In order to avoid the company between sample
Continuous property, improves the robustness of model, randomly selects 200 samples for training to every kind of fault type, remaining 100 sample is used
In test, therefore training set and test set separately include 2400 samples and 1200 samples in this example.
Referring to Fig. 2, the vibration time domain signal waveform of 12 kinds of different faults types of rolling bearing of the embodiment of the present invention, wherein scheming
2 (a) expression rolling bearings fault type be it is normal, Fig. 2 (b) expression rolling element failure, fault diameter be 0.007 inch, Fig. 2
(c) rolling element failure is indicated, fault diameter is 0.014 inch, and Fig. 2 (d) indicates rolling element failure, and fault diameter is 0.021 English
Very little, Fig. 2 (e) indicates inner ring failure, and fault diameter is 0.007 inch, and Fig. 2 (f) indicates inner ring failure, fault diameter 0.014
Inch, Fig. 2 (g) indicate inner ring failure, and fault diameter is 0.028 inch, and Fig. 2 (h) indicates outer ring failure, and fault diameter is
0.007 inch, failure orientation is direction of vertical 3 o'clock, and Fig. 2 (i) indicates outer ring failure, and fault diameter is 0.007 inch, failure
Orientation is 6 o'clock of level direction, and Fig. 2 (j) indicates outer ring failure, and fault diameter is 0.014 inch, and failure orientation is 6 points of level
Clock direction, Fig. 2 (k) indicate outer ring failure, and fault diameter is 0.021 inch, and failure orientation is direction of vertical 3 o'clock, Fig. 2 (l)
Indicate outer ring failure, fault diameter is 0.021 inch, and failure orientation is 6 o'clock of level direction.
Step 2) constructs N number of stacking-type self-encoding encoder model:
Constructed based on different activation primitive othernesses 5 respectively include 4 self-encoding encoders stacking-type self-encoding encoder model,
K-th of self-encoding encoder is denoted as in n-th of stacking-type self-encoding encoder model Hidden layer be5 stacking-types encode certainly
The number of nodes of the last one hidden layer of device model is 80, and the number of nodes of output layer is 12, wherein n=1,2 ..., 5, k
=1,2 ..., 4, the network structure of each stacking-type self-encoding encoder is as shown in table 1:
Table 1
Stacking-type self-encoding encoder number | K | Network structure |
1 | 4 | 400-200-100-80-12 |
2 | 4 | 400-200-100-80-12 |
3 | 4 | 400-200-100-80-12 |
4 | 4 | 400-200-100-80-12 |
5 | 4 | 400-200-100-80-12 |
Step 3) carries out parallel training to 5 stacking-type self-encoding encoder models:
Step 3a) enable n=1, k=1;
Step 3b) by training sample set X1As k-th of self-encoding encoder in n-th of stacking-type self-encoding encoder modelIt is defeated
Enter, it is rightIt is trained, the hidden layer after being trained isSelf-encoding encoder
Step 3c) by the self-encoding encoder after trainingHidden layerAs kth in n-th of stacking-type self-encoding encoder model
+ 1 self-encoding encoderInput, it is rightIt is trained, the hidden layer after being trained isSelf-encoding encoder
Step 3d) whether judge k=4 true, if so, n-th of stacking-type self-encoding encoder model after being trained, no
Then, k=k+1 is enabled, and executes step (3c)~(3d);
Step 3e) whether judge n=5 true, if so, the stacking-type self-encoding encoder model after 5 parallel trainings is obtained, it is no
Then, n=n+1 is enabled, and executes step (3b)~(3d);
Step 4) obtains multiple performance state features:
It will be in the stacking-type self-encoding encoder model after n-th of parallel training80 node output valves be used as from training sample
The n-th group performance state feature that this concentration is extracted, the performance state feature that 5 stacking-type self-encoding encoders are extracted is 400 altogether
A, m-th of performance state feature is Fm, m=1,2 ..., 400;
Step 5) is to each performance state feature FmCarry out characteristic evaluating:
Softmax disaggregated model is trained by minimizing error function, the softmax classification mould after being trained
Type, and by each performance state feature FmSoftmax disaggregated model after being input to training, obtains 400 Performance Evaluating Indexes
Value I1,I2,…,Im,…,I400;
Step 6) establishes the integrated multi-categorizer model based on softmax classifier:
Step 6a) by 20 iteration experiments, the relationship of Statistic analysis models classification diagnosis precision and threshold value is arranged 12
Threshold value is respectively 0.6,0.3,0.9,0.9,0.6,0.3,0.8,0.2,0.8,0.9,0.6,0.3, and 400 performance evaluations are referred to
Scale value one by one with q-th of threshold value TqIt is compared, filters out Performance Evaluating Indexes value greater than TqPerformance state feature composition characteristic
Subset Sq, obtain 12 character subset S1,S2,…,Sq,…,S12, q=1,2 ..., 12, the quantity of feature in each character subset
As shown in figure 3, character subset S1,S2,…,S12The quantity of middle performance state feature is respectively 185,122,20,9,34,69,32,
135、27、29、45、68。
Step 6b) by q-th of character subset SqIt is input to softmax classifier CqIn, obtain CqPrediction label vector
Rq,And calculate CqClassification diagnosis precision Acq, obtain the classification diagnosis precision Ac of 12 softmax classifiers1,
Ac2,…,Acq,…,Ac12;
Step 6c) setting threshold value Tc, by 12 classification diagnosis precision Ac1,Ac2,…,Acq,…,Ac12Respectively with TcCompare,
Classification diagnosis precision is filtered out greater than Tc10 classifier C1,C2,…,Cκ,…,C10, constitute based on softmax classifier
Integrated multi-categorizer MODEL C;
Step 7) obtains rolling bearing fault diagnosis result:
Step 7a) by X2In xjIt is input to each of C Cκ, κ=1,2 ..., 10, obtain 10 classifier C1,
C2,…,Cκ,…,C10To xjPrediction labelWherein,
Step 7b) pass through ballot method lookup prediction labelThe most pre- mark of middle frequency of occurrence
Label, and as the integrated multi-categorizer MODEL C based on softmax classifier to test sample xjPrediction label L(j):
Wherein, mode () intermediate scheme searches function,It indicates using the κ classifier filtered out to test specimens
This xjThe label of prediction;
Step 7c) according to prediction label L(j)Fault type belonging to each test sample is obtained, and then obtains rolling bearing
In the malfunction of different moments, the intelligent trouble diagnosis to rolling bearing is completed.Fig. 4 is to be tested using the present invention 1200
The result that sample data is classified, wherein in 100 test samples of the 2nd class, correct number of classifying is 92, is assigned to
The number of 3rd class is 1, and the number for being assigned to the 4th class is 6, and the number for being assigned to the 10th class is 1.
Below in conjunction with specific experiment, elaborate to technical effect of the invention.
1. experiment condition and content:
Central processing unit be Intel (R) Core (TM) i5-7500 3.40GHZ, memory 16G, WINDOWS7 operation system
On system, rolling bearing intelligent trouble diagnosis result is emulated with MATLAB R2017b software.
2. analysis of experimental results:
Application class diagnostic accuracy Acc of the present invention evaluates and tests the classification diagnosis precision of model, the expression formula of Acc are as follows:
In formula,L(j)For the label predicted j-th of test sample, y(j)Table
Show the physical tags of j-th of test sample.
Performance of the invention, specific comparative experiments are verified using two groups of comparative experimentss are as follows:
First group, the present invention is compared with the deep learning method based on single stacking-type self-encoding encoder model, it is right
Than the results are shown in Table 2:
Table 2
According to table 2 as can be seen that the classification diagnostic accuracy that proposes is 95.42% in the present invention, i.e., wherein 1145
Test sample classification is accurate.Compared with the deep learning method based on single stacking-type self-encoding encoder model, classification diagnosis essence
Degree is significantly increased.
Second group, the present invention is compared with the intelligent failure diagnosis method based on single sorter model, comparison knot
Fruit is shown in Table 3:
Table 3
According to table 3 as can be seen that due to having carried out characteristic evaluating and screening to the feature of extraction, and it is based on single heap
Stack self-encoding encoder model is compared, and the classification diagnosis precision based on single classifier is improved on the whole.But most accurately
Single classifier method low 5% of the precision still than being proposed in the present invention.
In conclusion the present invention can filter out and diagnose target correlation strong, more representative feature and overcome list
The deficiency of a softmax classifier poor robustness, improves the precision of rolling bearing intelligent trouble diagnosis.
Claims (4)
1. a kind of rolling bearing intelligent failure diagnosis method based on Concurrent Feature study and integrated multi-categorizer, it is characterised in that
Include the following steps:
(1) training sample set and test sample collection are obtained:
(1a) is using I vibration time-domain signal data of the rolling bearing chosen from database as training sample, each training
Sample includes a kind of label for indicating fault category, and the classification sum of label is Q, all training sample composing training sample sets
X1,Wherein, I >=2000, and I > > Q, xiIndicate i-th of training sample, y(i)Indicate xiLabel;
(1b) is using J vibration time-domain signal data of the rolling bearing to be diagnosed acquired in real time by data collection system as survey
Sample sheet, all test samples constitute test sample collection X2,J≥I/2,xjIndicate j-th of test sample;
(2) N number of stacking-type self-encoding encoder model is constructed:
Constructed based on different activation primitive othernesses it is N number of respectively include K self-encoding encoder stacking-type self-encoding encoder model, n-th
K-th of self-encoding encoder is denoted as in a stacking-type self-encoding encoder modelHidden layer beN number of stacking-type encodes certainly
The number of nodes of the last one hidden layer of device model is h, and the number of nodes of output layer is o, wherein n=1,2 ..., N, k=
1,2 ..., K, N >=2, K >=2;
(3) parallel training is carried out to N number of stacking-type self-encoding encoder model:
(3a) enables n=1, k=1;
(3b) is by training sample set X1As k-th of self-encoding encoder in n-th of stacking-type self-encoding encoder modelInput, it is right
It is trained, the hidden layer after being trained isSelf-encoding encoder
(3c) is by the self-encoding encoder after trainingHidden layerCertainly as kth+1 in n-th of stacking-type self-encoding encoder model
EncoderInput, it is rightIt is trained, the hidden layer after being trained isSelf-encoding encoder
Whether (3d) judges k=K true, if so, otherwise n-th of stacking-type self-encoding encoder model after being trained enables k=k
+ 1, and execute step (3c)~(3d);
Whether (3e) judges n=N true, if so, otherwise obtaining the stacking-type self-encoding encoder model after N number of parallel training enables n
=n+1, and execute step (3b)~(3d);
(4) multiple performance state features are obtained:
It will be in the stacking-type self-encoding encoder model after n-th of parallel trainingH node output valve be used as from training sample set
In the n-th group performance state feature extracted, the performance state feature that all stacking-type self-encoding encoders are extracted is M altogether,
In, M=N × h, m-th of performance state feature is Fm, m=1,2 ..., M;
(5) to each performance state feature FmCarry out characteristic evaluating:
Softmax disaggregated model is trained by minimizing error function, the softmax disaggregated model after being trained,
And by each performance state feature FmSoftmax disaggregated model after being input to training, obtains M performance evaluation index value I1,
I2,…,Im,…,IM;
(6) the integrated multi-categorizer model based on softmax classifier is established:
Q threshold value T is arranged in (6a)1,T2,…,Tq,…,TQ, and by M performance evaluation index value one by one with q-th of threshold value TqIt carries out
Compare, filters out Performance Evaluating Indexes value greater than TqPerformance state feature composition characteristic subset Sq, obtain Q character subset S1,
S2,…,Sq,…,SQ, q=1,2 ..., Q;
(6b) is by q-th of character subset SqIt is input to softmax classifier CqIn, obtain CqPrediction label vector Rq,And calculate CqNicety of grading Acq, obtain the nicety of grading Ac of Q softmax classifier1,Ac2,…,Acq,…,
AcQ;
Threshold value T is arranged in (6c)c, by Q nicety of grading Ac1,Ac2,…,Acq,…,AcQRespectively with TcCompare, filters out classification essence
Degree is greater than Tcλ classifier C1,C2,…,Cκ,…,Cλ, λ >=5 constitute the integrated multi-categorizer mould based on softmax classifier
Type C;
(7) rolling bearing fault diagnosis result is obtained:
(7a) is by X2In xjIt is input to each of C Cκ, κ=1,2 ..., λ obtain λ classifier C1,C2,…,Cκ,…,Cλ
To xjPrediction labelWherein,
(7b) searches prediction label by ballot methodThe most prediction label of middle frequency of occurrence, and by its
As the integrated multi-categorizer MODEL C based on softmax classifier to test sample xjPrediction label L(j):
Wherein, mode () intermediate scheme searches function,It indicates using the κ classifier filtered out to test sample xjIn advance
The label of survey, λ indicate the number of the classifier filtered out;
(7c) is by prediction label L(j)The fault category for including with training sample set is mapped, and obtains rolling bearing when different
The malfunction at quarter.
2. the rolling bearing intelligent trouble diagnosis according to claim 1 based on Concurrent Feature study and integrated multi-categorizer
Method, which is characterized in that minimum error function described in step (5), expression formula are as follows:
Wherein, y(i)Indicate that the fault category label that i-th of training sample includes, I indicate that training sample concentrates the total of training sample
Number, q indicate that q class fault category, Q indicate fault category sum, FmIndicate m-th of feature vector, λqIndicate softmax classification
Model corresponds to the weight of q class failure, offset parameter vector, and D { } indicates target function,
3. the rolling bearing intelligent trouble diagnosis according to claim 1 based on Concurrent Feature study and integrated multi-categorizer
Method, which is characterized in that Performance Evaluating Indexes value I described in step (5)m, expression formula are as follows:
Wherein, hλ(Fm) indicate that softmax model is based on feature FmOutput label, q indicate q class fault category, Q indicate therefore
Hinder classification sum, label indicates that training sample label, num () indicate counting function.
4. the rolling bearing intelligent trouble diagnosis according to claim 1 based on Concurrent Feature study and integrated multi-categorizer
Method, which is characterized in that calculating C described in step (6b)qNicety of grading AcqCalculation formula are as follows:
Wherein,I indicates that the label of i-th of training sample, I indicate training sample set
The number of middle training sample, Rq (i)It indicates to utilize softmax classifier CqTo the label of i-th of training sample prediction, y(i)It indicates
The fault category label that i-th of training sample includes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910331462.4A CN110110768B (en) | 2019-04-24 | 2019-04-24 | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910331462.4A CN110110768B (en) | 2019-04-24 | 2019-04-24 | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110110768A true CN110110768A (en) | 2019-08-09 |
CN110110768B CN110110768B (en) | 2021-02-26 |
Family
ID=67486436
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910331462.4A Active CN110110768B (en) | 2019-04-24 | 2019-04-24 | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110110768B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110689072A (en) * | 2019-09-25 | 2020-01-14 | 重庆大学 | Knowledge transfer-based dynamic industrial data classification method |
CN110826607A (en) * | 2019-10-24 | 2020-02-21 | 北京建筑大学 | Fault detection method and device for rolling bearing |
CN110909826A (en) * | 2019-12-10 | 2020-03-24 | 新奥数能科技有限公司 | Diagnosis monitoring method and device for energy equipment and electronic equipment |
CN111680665A (en) * | 2020-06-28 | 2020-09-18 | 湖南大学 | Motor mechanical fault diagnosis method based on data driving and adopting current signals |
CN112052148A (en) * | 2020-08-17 | 2020-12-08 | 烽火通信科技股份有限公司 | Fault prediction method, device, equipment and readable storage medium |
CN112069621A (en) * | 2020-09-08 | 2020-12-11 | 西安电子科技大学 | Method for predicting residual service life of rolling bearing based on linear reliability index |
CN112327219A (en) * | 2020-10-29 | 2021-02-05 | 国网福建省电力有限公司南平供电公司 | Distribution transformer fault diagnosis method with automatic feature mining and automatic parameter optimization |
CN112949591A (en) * | 2021-03-31 | 2021-06-11 | 上海电力大学 | Rolling bearing fault diagnosis method based on depth self-encoder |
CN115407753A (en) * | 2022-08-18 | 2022-11-29 | 广东元梦泽技术服务有限公司 | Industrial fault diagnosis method for multivariate weighted ensemble learning |
CN117359391A (en) * | 2023-12-08 | 2024-01-09 | 江苏雷鸣智能装备有限公司 | Intelligent fault diagnosis method and system for rolling bearing of numerical control machine tool |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001096599A2 (en) * | 2000-06-15 | 2001-12-20 | Philogen S.R.L. | Methods for quantitative determination of b-fibronectin in biological fluids and tissues |
CN105758644A (en) * | 2016-05-16 | 2016-07-13 | 上海电力学院 | Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy |
CN106124212A (en) * | 2016-06-16 | 2016-11-16 | 燕山大学 | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
CN109211546A (en) * | 2018-08-28 | 2019-01-15 | 电子科技大学 | Rotary machinery fault diagnosis method based on noise reduction autocoder and incremental learning |
-
2019
- 2019-04-24 CN CN201910331462.4A patent/CN110110768B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001096599A2 (en) * | 2000-06-15 | 2001-12-20 | Philogen S.R.L. | Methods for quantitative determination of b-fibronectin in biological fluids and tissues |
CN105758644A (en) * | 2016-05-16 | 2016-07-13 | 上海电力学院 | Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy |
CN106124212A (en) * | 2016-06-16 | 2016-11-16 | 燕山大学 | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine |
CN106682688A (en) * | 2016-12-16 | 2017-05-17 | 华南理工大学 | Pile-up noise reduction own coding network bearing fault diagnosis method based on particle swarm optimization |
CN109211546A (en) * | 2018-08-28 | 2019-01-15 | 电子科技大学 | Rotary machinery fault diagnosis method based on noise reduction autocoder and incremental learning |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110689072A (en) * | 2019-09-25 | 2020-01-14 | 重庆大学 | Knowledge transfer-based dynamic industrial data classification method |
CN110689072B (en) * | 2019-09-25 | 2023-04-07 | 重庆大学 | Knowledge transfer-based dynamic industrial data classification method |
CN110826607A (en) * | 2019-10-24 | 2020-02-21 | 北京建筑大学 | Fault detection method and device for rolling bearing |
CN110909826A (en) * | 2019-12-10 | 2020-03-24 | 新奥数能科技有限公司 | Diagnosis monitoring method and device for energy equipment and electronic equipment |
CN111680665A (en) * | 2020-06-28 | 2020-09-18 | 湖南大学 | Motor mechanical fault diagnosis method based on data driving and adopting current signals |
CN112052148B (en) * | 2020-08-17 | 2022-04-29 | 烽火通信科技股份有限公司 | Fault prediction method, device, equipment and readable storage medium |
CN112052148A (en) * | 2020-08-17 | 2020-12-08 | 烽火通信科技股份有限公司 | Fault prediction method, device, equipment and readable storage medium |
CN112069621A (en) * | 2020-09-08 | 2020-12-11 | 西安电子科技大学 | Method for predicting residual service life of rolling bearing based on linear reliability index |
CN112327219A (en) * | 2020-10-29 | 2021-02-05 | 国网福建省电力有限公司南平供电公司 | Distribution transformer fault diagnosis method with automatic feature mining and automatic parameter optimization |
CN112327219B (en) * | 2020-10-29 | 2024-03-12 | 国网福建省电力有限公司南平供电公司 | Distribution transformer fault diagnosis method with automatic feature mining and parameter automatic optimizing functions |
CN112949591A (en) * | 2021-03-31 | 2021-06-11 | 上海电力大学 | Rolling bearing fault diagnosis method based on depth self-encoder |
CN115407753A (en) * | 2022-08-18 | 2022-11-29 | 广东元梦泽技术服务有限公司 | Industrial fault diagnosis method for multivariate weighted ensemble learning |
CN115407753B (en) * | 2022-08-18 | 2024-02-09 | 广东元梦泽技术服务有限公司 | Industrial fault diagnosis method for multi-variable weighting integrated learning |
CN117359391A (en) * | 2023-12-08 | 2024-01-09 | 江苏雷鸣智能装备有限公司 | Intelligent fault diagnosis method and system for rolling bearing of numerical control machine tool |
CN117359391B (en) * | 2023-12-08 | 2024-02-13 | 江苏雷鸣智能装备有限公司 | Intelligent fault diagnosis method and system for rolling bearing of numerical control machine tool |
Also Published As
Publication number | Publication date |
---|---|
CN110110768B (en) | 2021-02-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110110768A (en) | Fault Diagnosis of Roller Bearings based on Concurrent Feature study and multi-categorizer | |
CN109781411B (en) | Bearing fault diagnosis method combining improved sparse filter and KELM | |
CN109186973B (en) | Mechanical fault diagnosis method of unsupervised deep learning network | |
CN106769052B (en) | A kind of mechanical system rolling bearing intelligent failure diagnosis method based on clustering | |
CN110334764A (en) | Rotating machinery intelligent failure diagnosis method based on integrated depth self-encoding encoder | |
CN109782603A (en) | The detection method and monitoring system of rotating machinery coupling fault | |
CN109460618A (en) | A kind of rolling bearing remaining life on-line prediction method and system | |
CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
CN111460728B (en) | Method and device for predicting residual life of industrial equipment, storage medium and equipment | |
CN106874957A (en) | A kind of Fault Diagnosis of Roller Bearings | |
CN109655259A (en) | Combined failure diagnostic method and device based on depth decoupling convolutional neural networks | |
CN112765890B (en) | Dynamic domain adaptive network-based multi-working-condition rotating machine residual life prediction method | |
CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
CN109102032A (en) | A kind of pumping plant unit diagnostic method based on depth forest and oneself coding | |
CN112308147B (en) | Rotary machinery fault diagnosis method based on multi-source domain anchor adapter integrated migration | |
CN108106844B (en) | A kind of Method for Bearing Fault Diagnosis of the automatic coding machine of self-tuning parameter adjustment | |
CN206504869U (en) | A kind of rolling bearing fault diagnosis device | |
CN108256556A (en) | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network | |
CN110132554A (en) | A kind of rotary machinery fault diagnosis method that depth Laplce encodes certainly | |
CN106022352A (en) | Submersible piston pump fault diagnosis method based on support vector machine | |
CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
CN110348468A (en) | A kind of bearing inferior health recognition methods of the strong reconstruct edge noise reduction autocoder of Method Using Relevance Vector Machine optimization | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN111783941B (en) | Mechanical equipment diagnosis classification method based on probability confidence convolutional neural network | |
CN113505639A (en) | TPE-XGboost-based rotating machine multi-parameter health state evaluation 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 |