CN101886984A - Method for railway steel rail stress real-time detection and classification - Google Patents
Method for railway steel rail stress real-time detection and classification Download PDFInfo
- Publication number
- CN101886984A CN101886984A CN2009100511398A CN200910051139A CN101886984A CN 101886984 A CN101886984 A CN 101886984A CN 2009100511398 A CN2009100511398 A CN 2009100511398A CN 200910051139 A CN200910051139 A CN 200910051139A CN 101886984 A CN101886984 A CN 101886984A
- Authority
- CN
- China
- Prior art keywords
- classification
- rail stress
- support vector
- vector machine
- diagnosis
- 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.)
- Pending
Links
Images
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention is used in the field of railway steel rail stress identification and detection. In the invention, by classifying the rail stress data detected by a detector, a stress concentration condition is determined. The method comprises: firstly, performing fault diagnosis by using a neural network according to an iron rail stress condition, and realizing the visualization of the system; and secondly, performing fault pre-warning diagnosis by continuously testing new data and introducing a support vector machine-based method. In the invention, a 'paired classification' multi-class classification algorithm and a new M-ary support vector machine multi-class classification algorithm are adopted to realize the classification of the various faults, and are introduced into the iron rail stress fault diagnosis, and thus, the classification and fault diagnosis effects are desirable.
Description
One, technical field
The present invention utilizes support vector machine method that rail stress is carried out fault diagnosis, the data that detecting instrument records classified, thus the concentrated situation of judgement stress.Adopt new " classification in pairs " that the multiclass fault is classified, by test shows, classifying quality is very good, and this proves that also support vector machine is to the fault diagnosis application success in the actual engineering of classifying.Use LIBSVM software to classify, utilize SQL Server 2000 and VC++ software engineering to realize the visual of system.
Two, background technology
Fault diagnosis technology is an applied interdisciplinary study, and its theoretical foundation relates to modern control theory, computer engineering, mathematical statistics, signal Processing, pattern-recognition, artificial intelligence and application corresponding subject.Fault diagnosis technology mainly comprises the content of three aspects: fault detect, fault isolation, fault identification.According to the sorting technique of routine the method for fault diagnosis can be divided into method based on analytic model, based on method for processing signals with based on the method three major types of knowledge.Comprise three kinds of parameter estimation method, state estimation method and equivalent space methods based on the method for diagnosing faults of analytic model; Method for diagnosing faults based on signal Processing; Method for diagnosing faults based on knowledge mainly can be divided into: expert system method for diagnosing faults, fuzzy fault diagnostic method, fault tree method for diagnosing faults, neural network failure diagnostic method, the method for diagnosing faults based on support vector machine, diagnosing information fusion fault method and based on the Agent method for diagnosing faults.
Method of the present invention is based on the method for diagnosing faults of support vector machine.
A kind of novel machine learning method---the support vector machine (SVM, support vector machine) that Vapnik proposed in nineteen ninety-five.Support vector machine has complete Statistical Learning Theory basis and outstanding learning performance, has become the new focus of research of machine learning circle, and has all obtained successful application in a lot of fields.
SVM is the binary classification device, and SVM is applied on the fault diagnosis, must be by the SVM structure multivariate classification device of binary, i.e. multicategory classification device.Mainly comprise four steps: the first, data initialization.In order to reduce the influence that different parameters absolute figure size is diagnosed support vector machine in the fault sample, need carry out the normalization pre-service to each parameter dimension of sample.The second, sample learning is learnt fault sample by sorter, obtains the optimum lineoid of different faults type.The 3rd, Fault Identification is imported new state sample in sorter, and system is carried out Fault Identification.The 4th, new fault handling when new failure mode and state appear in system, can be input in the sorter, obtains new optimum lineoid, improves the Tracking Recognition ability of fault diagnosis.
Three, summary of the invention
The present invention is directed to railroad track stress and detect, the data that detecting instrument records are classified, thereby judge the concentrated situation of rail stress.Development one cover railroad track trouble stress detects intelligence system, and Fig. 1 and 2 is R﹠D process and raw data.
The particular content of invention is as follows:
1) neural network is carried out fault diagnosis to the rail stress state, and the system of realization is visual.
Carrying out the initial stage of problem work, at first select for use common BP neural network that data are classified.Diagnostic techniques scheme based on the BP neural network: comprise the 1. preparation of sample data collection; 2. network structure design; 3. network training and test, the network training error that obtains as shown in Figure 3.
System's The Visual Implementation: the native system database adopts SQL Server 2000, and programming adopts Visual C++ software to realize functions such as the calling of the design of visualization interface, database, graphic presentation.System realizes that thinking is: 1. carry out neural metwork training under MATLAB 6.5 environment, the training data of handling well is sent into the BP neural network, obtain parameters such as optimum weights and threshold value after the extraction training, it is kept in the notepad.2. under Visual C++ environment, data such as the weights preserved in advance, threshold value are read in, utilize test sample book to calculate then by network structure, and the output result.
2) find that by the test new data generalization ability of neural network model is not very good, and then introduce method for diagnosing faults based on support vector machine.
The initial stage is carried out in subject study, has adopted the BP neural network that iron staff stress test data are classified, and when testing, discrimination is 88.89% with 18 groups of samples (i.e. 360 data).As seen, the model that obtains with neural network, though be that the precision of error in classification can reach very high, its generalization ability still remains to be improved, this can find out from the discrimination of network model to unknown sample.Therefore adopt support vector machine that iron staff pressure test data are classified in the hope of obtaining higher discrimination.
Multiclass Classification adopts classification in pairs.With three kinds of stress state numberings, i.e. 0 representative is normal, and on behalf of stress, 1 concentrate, and 2 represent hurt.Need set up three sorters, i.e. SVM01, SVM02 and SVM03 according to the data pattern classification.Kernel function is selected RBF nuclear, the parameter selection tool grid.py that carries according to the LIBSVM tool box of the selection of parameter σ and penalty factor C wherein, and these two parameters are represented with g and C respectively in LIBSVM.Training data all adopts the data identical with the BP neural network with test data, and promptly training data is that every kind of mode data is 50 groups, and test data is 9 groups of 0 classes (normally), 3 groups of 1 classes (stress is concentrated), 6 groups of 2 classes (hurt).Classification results sees Table 1.
Table 1: each sorter classification results
Obviously, according to " maximum ballot method " as can be seen from the table, all obtained correct identification for test data according to the output of 3 sorters.Obviously effect has proved SVM generalization ability preferably than the discrimination height of BP neural network to test sample book.
3) with a kind of new Multiclass Classification---M-ary support vector machine sorting algorithm is incorporated in the rail stress failures diagnosis, finds that this algorithm when guaranteeing classifying quality, compares with paired sorter, and the sorter number that needs to set up obtains reducing.
M-ary support vector machine sorting algorithm has not only made full use of the advantage of two-value sorter: do not rely on priori, calculate simply relatively, and only need construct when handling K class problem
Individual sorter implements more simple and convenient.Its realization principle is seen Fig. 4.
M-ary support vector machine sorting algorithm has than tradition and manys the apparent in view several advantages of sorting algorithm.The first, the method makes full use of the classification capacity of two-value SVM, sets up a plurality of non-linear classification boundaries faces by the multiclass sample set being split into two classes, utilizes these boundary surfaces to separate more classification then.The second, M-ary support vector machine sorting algorithm only needs
Sorter is o (K) or o (K yet traditional linear multi-categorizer based on the border needs the complexity of sorter number usually
2).The 3rd, the M-ary algorithm of support vector machine is a kind of very simple algorithm, and is the same with the two-value sorter, and it also is that data rely on, and do not need priori with the data of statistical dependence.
Using the M-ary sorting algorithm to carry out the rail stress failures diagnoses idiographic flow as follows:
1. according to 2 classifications that sorter divided, training sample is split, and as for sorter 1, positive class sample is made of the data of stress collected state in the original training data, negative class sample is made of the data of normal and hurt two states, and the training data of sorter 2 constitutes similarly;
2. utilize the training data of ready two sorters in the step 1) that it is trained, obtain corresponding disaggregated model;
3. test data is sent to respectively in sorter 1 and the sorter 2, obtains output classification separately;
4. by formula
Calculate the affiliated final classification of checking sample.
The emulation experiment situation sees Table 2 and table 3.
Table 2: simulation result
This shows, use the M-ary sorting technique, the classification results that obtains and 2) in pairs the nicety of grading of sorter can compare.
In this section, data category is 3 classes, by using the M-ary sorting technique sorter number that paired sorter uses is reduced to 2 by 3.To 7 class transformer fault diagnosis, utilization M-ary sorting technique can realize by 3 sorters, if then need 7* (7-1)=21 sorter by paired sorter.So under a lot of situation of classification, the advantage of the method is fairly obvious.
Claims (4)
1. the present invention is directed to that railway steel rail stress is classified, failure prediction and diagnosis.
2. on the basis of claim (1), propose new " classification in pairs " the multiclass fault is carried out category forecast and diagnosis.
3. with neural network the rail stress state is carried out category forecast and diagnosis.
4. utilize the M-ary algorithm of support vector machine that railway steel rail stress is carried out category forecast and diagnosis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100511398A CN101886984A (en) | 2009-05-13 | 2009-05-13 | Method for railway steel rail stress real-time detection and classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009100511398A CN101886984A (en) | 2009-05-13 | 2009-05-13 | Method for railway steel rail stress real-time detection and classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101886984A true CN101886984A (en) | 2010-11-17 |
Family
ID=43072989
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009100511398A Pending CN101886984A (en) | 2009-05-13 | 2009-05-13 | Method for railway steel rail stress real-time detection and classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101886984A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103359137A (en) * | 2012-03-31 | 2013-10-23 | 上海申通地铁集团有限公司 | Turnout fault early warning method |
| CN106980815A (en) * | 2017-02-07 | 2017-07-25 | 王俊 | Facial paralysis objective evaluation method under being supervised based on H B rank scores |
-
2009
- 2009-05-13 CN CN2009100511398A patent/CN101886984A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103359137A (en) * | 2012-03-31 | 2013-10-23 | 上海申通地铁集团有限公司 | Turnout fault early warning method |
| CN106980815A (en) * | 2017-02-07 | 2017-07-25 | 王俊 | Facial paralysis objective evaluation method under being supervised based on H B rank scores |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109952614B (en) | Biological particle classification system and method | |
| CN105653444B (en) | Software defect fault recognition method and system based on internet daily record data | |
| CN101750210B (en) | Fault diagnosis method based on OLPP feature reduction | |
| CN103914064B (en) | Based on the commercial run method for diagnosing faults that multi-categorizer and D-S evidence merge | |
| CN114048568B (en) | Rotary machine fault diagnosis method based on multisource migration fusion shrinkage framework | |
| CN109447461B (en) | User credit evaluation method and device, electronic equipment and storage medium | |
| CN106973057A (en) | A kind of sorting technique suitable for intrusion detection | |
| CN102498445A (en) | Supervised fault learning using rule-generated samples for machine condition monitoring | |
| Shen et al. | A deep multi-label learning framework for the intelligent fault diagnosis of machines | |
| CN101738998B (en) | System and method for monitoring industrial process based on local discriminatory analysis | |
| CN111580506A (en) | Industrial process fault diagnosis method based on information fusion | |
| CN106482967A (en) | A kind of Cost Sensitive Support Vector Machines locomotive wheel detecting system and method | |
| Xing et al. | Intelligent fault diagnosis of rolling bearing based on novel CNN model considering data imbalance | |
| CN108038493A (en) | Based on the industrial big data procedure fault monitoring method for returning intelligent distinguishing insertion | |
| CN105718937A (en) | Multi-class object classification method and system | |
| CN116894187A (en) | Gear box fault diagnosis method based on deep migration learning | |
| CN109523514A (en) | To the batch imaging quality assessment method of Inverse Synthetic Aperture Radar ISAR | |
| CN109254577A (en) | A kind of intelligence manufacture procedure fault classification method and device based on deep learning | |
| CN115165366A (en) | Variable working condition fault diagnosis method and system for rotary machine | |
| CN114997211A (en) | Cross-working-condition fault diagnosis method based on improved countermeasure network and attention mechanism | |
| CN114462475A (en) | Unsupervised machine abnormal sound detection method and unsupervised machine abnormal sound detection device based on single classification algorithm | |
| CN113884300A (en) | Rolling bearing fault diagnosis method for deep anti-migration learning | |
| CN101886984A (en) | Method for railway steel rail stress real-time detection and classification | |
| CN113221946A (en) | Method for diagnosing fault types of mechanical equipment | |
| CN104537392A (en) | Object detection method based on distinguishing semantic component learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20101117 |