CN103728135A - Bearing fault feature extraction and diagnosis method of non-negative matrix factorization - Google Patents
Bearing fault feature extraction and diagnosis method of non-negative matrix factorization Download PDFInfo
- Publication number
- CN103728135A CN103728135A CN201310693212.8A CN201310693212A CN103728135A CN 103728135 A CN103728135 A CN 103728135A CN 201310693212 A CN201310693212 A CN 201310693212A CN 103728135 A CN103728135 A CN 103728135A
- Authority
- CN
- China
- Prior art keywords
- bearing
- train
- fault
- signal
- matrix factorization
- 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
Abstract
Provided is a bearing fault feature extraction and diagnosis method of non-negative matrix factorization. Bearing vibration signal collection is conducted on an acceleration sensor, amplitude spectrums corresponding to the acceleration sensor are obtained through short-time Fourier transformation, and the amplitude spectrums are selected randomly, vectored, and recombined to form training set matrixes. Due to the fact that the non-negative matrix factorization multiplying iterative algorithm based on the Euclidean distance is conducted on the training set matrixes, basis matrixes containing fault characteristic quantities can be extracted. The basis matrixes are combined, and over-complete characteristic atom dictionaries with respect to a bearing can be obtained. Finally, the amplitude spectrums of test signals are projected to the characteristic atom dictionaries, and fault types of the test signals are judged according to the maximum value of each line (row) in encoding matrixes. The bearing fault feature extraction and diagnosis method reduces difficulty of the fault characteristic algorithm of the bearing, improves operation efficiency, and improves the recognition effect.
Description
Technical field
The present invention relates to bearing typical fault feature, diagnostic techniques field, the bearing fault characteristics that is specifically related to a kind of Non-negative Matrix Factorization extracts and diagnostic method.
Background technology
Bearing is as one of part of the most general in modern machinery and equipment, most critical, and its running status will play conclusive effect to the usability of whole equipment.So, find that as early as possible the fault of generation can be avoided many serious consequences.As one of most widely used method, vibration analysis can, under frequency domain, judges running status and identify fault, but existing method often need higher specialty background from vibration signal.Therefore, in the past few decades, the technology that are easy to promote and possess stable evaluation result become the emphasis of research gradually, and, many intelligent methods have obtained application in commercial production, have improved greatly automaticity and the accuracy of diagnosis.
Feature extraction is regarded as a most important ring in diagnostic process, and suitable treatment technology can accurately extract the characteristic component relevant to the running status of component of machine.But the extraction and analytical method Computing Principle of main flow is complicated at present, result physical significance is not strong, and accuracy rate is not high.How optimized calculation method improve accuracy rate, is still good problem in the urgent need to address in failure diagnostic process.
In order effectively to extract the Weak characteristic amount in bearing vibration signal, Non-negative Matrix Factorization provides a kind of effective means,, by the non-negative raw data matrix decomposing, realizes nonlinear signal characteristic abstraction.Yet in actual measurement, vibration signal is mostly the one-dimensional signal that contains positive negative value, cannot meet the treatment conditions of Non-negative Matrix Factorization simultaneously, therefore, consider first gathered signal to be transformed under time-frequency domain.And in the identification of signal after obtaining characteristic quantity, current research mainly adopts the methods such as support vector machine or artificial neural network, is still having very large optimization prospect aspect counting yield and accuracy.And the assembility that makes full use of Non-negative Matrix Factorization method realize fault-signal automatically the technology of identification be not also applied.
Summary of the invention
In order to solve above-mentioned technical deficiency, the object of the present invention is to provide a kind of bearing fault characteristics of Non-negative Matrix Factorization to extract and diagnostic method, solve feature extraction and pattern clustering problem for bearing fault signal, thereby realized the automatic identification of fault mode.
To achieve the above object, the technical solution adopted in the present invention is:
The bearing fault characteristics of Non-negative Matrix Factorization extracts and a diagnostic method, comprises the following steps:
1) utilize acceleration transducer to gather the vibration signal x of rolling bearing
i(n), n is sampling number, and n is greater than zero integer, and i is signal numbering;
2) utilize Short Time Fourier Transform, calculate vibration signal x
i(n) time-frequency distributions S
i, wherein slippage window is long for [1/2
3* n, 1/2*n];
3) to the time-frequency distributions S obtaining through conversion
idelivery, only retains amplitude information | S
i|, and by all amplitude spectrums | S
i| matrix-vector;
4), from i class signal, choose arbitrarily the amplitude spectrum composing training collection after a part of vectorization
V
train(i)={υec(|S
1|),υec(|S
2|),...,υec(|S
t|)} (1)
5) according to the definition of Non-negative Matrix Factorization
To V
train(i) decompose, obtain
V
train(i)≈W
n×r(i)H
r×m(i) (3)
W wherein
n * r(i) be basis matrix, be considered as the feature of extraction, and H
r * m(i) be weight coefficient matrix;
6) by the basis matrix W of various types of signal
n * r(i) close the over-complete dictionary of atoms W that is arranged as a rolling bearing fault
train,
W
train={W
n×r(1),W
n×r(2),...,W
n×r(T)} (6)
7) extract vibration-testing signal y (n), by Short Time Fourier Transform, obtain the time-frequency spectrum that only comprises amplitude information | Y|;
8) by amplitude spectrum | Y| is to former word bank W of training stage
trainprojection, according to the Clustering features of Non-negative Matrix Factorization, realizes Fault Identification.
Because the present invention takes full advantage of the Clustering features of Non-negative Matrix Factorization, the former word bank projection by obtaining to the training stage, has realized the automatic identification of bearing various faults signal.Compared with the existing methods, not only utilized fully the data of training stage, improved operation efficiency, and recognition accuracy has also made moderate progress.
Accompanying drawing explanation
Fig. 1 is process flow diagram of the present invention.
Fig. 2 is the vibration signal of rolling body fault in the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils.
Fig. 3 is the interior ring fault vibration signal of the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils.
Fig. 4 is the outer shroud fault vibration signal of the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils
Fig. 5 is the amplitude spectrum of the rolling body fault of the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils.
Fig. 6 is the amplitude spectrum of the interior ring fault of the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils.
Fig. 7 is the amplitude spectrum of the outer shroud fault of the embodiment of the present invention: (a) 7mils wherein, (b) 14mils, (c) 21mils.
Fig. 8 is the recognition result of test sample book.
Embodiment
Below in conjunction with accompanying drawing, the present invention is described in detail.
With reference to figure 1, a kind of bearing fault characteristics of Non-negative Matrix Factorization extracts and diagnostic method, comprises the following steps:
1) utilize acceleration transducer to gather the vibration signal x of each class bearing
i(n), n is sampling number, and n is greater than zero integer, and i is signal numbering;
2) utilize Short Time Fourier Transform, calculate vibration signal x
i(n) time-frequency distributions S
i, wherein slippage window is long for [1/2
3* n, 1/2*n];
3) to the time-frequency distributions S obtaining through conversion
idelivery, only retains amplitude information | S
i|, and by all amplitude spectrums | S
i| matrix-vector;
4), from i class signal, choose arbitrarily the amplitude spectrum composing training collection after a part of vectorization
V
train(i)={υec(|S
1|),υec(|S
2|),...,υec(|S
t|)} (1)
5) according to the definition of Non-negative Matrix Factorization
To V
train(i) decompose, obtain
V
train(i)≈W
n×r(i)H
r×m(i) (3)
W wherein
n * r(i) be basis matrix, be considered as the feature of extraction, and H
r * m(i) be weight coefficient matrix;
5.1) select with reference to the target formula (2) based on Euclidean distance definition;
5.2) adopt PCA method, the dimension r that selected characteristic value accounts for total amount 90% is dimension to be decomposed;
5.3) random initializtion basis matrix W and weight matrix H;
5.4) according to decomposing requirement, iteration dimension or error size are set, by (4), (5) iteration
6) by the basis matrix W of various types of signal
n * r(i) assembled arrangement is the over-complete dictionary of atoms W of a rolling bearing typical fault
train,
W
train={W
n×r(1),W
n×r(2),...,W
n×r(T) (6)
7), to any vibration-testing signal y (n), by Short Time Fourier Transform, obtain the time-frequency spectrum that only comprises amplitude information | Y|;
8) by amplitude spectrum | after Y| vectorization, the former word bank W obtaining to the training stage
trainprojection; According to the Clustering features of Non-negative Matrix Factorization, the weight coefficient matrix H that projection is obtained is considered as coded message, characterizes W
trainin the contribution of each atom pair test signal, in H, element numerical value shows that more greatly the contribution of its corresponding atom is larger, just can think that test signal belongs to the corresponding fault type of this type of atom, suc as formula (7)
k=argmax[H
.j] (7)
Below in conjunction with instantiation, the present invention is described in detail.
First, in the various faults vibration-testing of SKF6205 deep groove ball bearing, adopt acceleration transducer to gather the vibration signal x (n) of various states lower bearing, wherein, systematic sampling frequency 12000Hz, get sampling number n=1024, Fig. 2, Fig. 3, Figure 4 shows that the vibration signal time domain waveform of various states lower bearing;
Secondly, adopt the Hanning window Short Time Fourier Transform that window length is 128, and only retain amplitude information by delivery operator, the amplitude spectrum result obtaining is as shown in Fig. 5, Fig. 6, Fig. 7.
Then, will randomly draw a part to various types of signal, and by vectorization composing training collection, then adopt Non-negative Matrix Factorization method to be decomposed training set, obtain the characteristic information about all kinds of vibration signals,
1), according to PCA method, determine that original amplitude spectrum dimension to be decomposed is 6;
2) the target formula (2) based on Euclidean distance according to Non-negative Matrix Factorization, by formula (4), (5) iteration 500 times, decomposes and obtains basis matrix W and weight matrix H;
3) the basis matrix W that contains characteristic information of all kinds of decomposition is merged into the over-complete dictionary of atoms of fault bearing;
Finally, by the former word bank projection to the training stage by the test signal that contains 9 class amplitude spectrums, according to the position at the maximal value place of each row in weight coefficient matrix H, in conjunction with former word bank W
traininformation, judgement test signal associative mode classification, realizes the automatic identification of fault mode, result as shown in Figure 8.According to experimental result repeatedly, find the fault type of the multiple unknown signaling of the method energy Division identification.And contrast with the recognition methods based on artificial neural network, verified the validity of invention.
By above application note, the present invention is based on Non-negative Matrix Factorization method bearing fault characteristics extraction and diagnostic method have not only been optimized to fault recognition rate, and algorithm easy to understand, take full advantage of training stage result of calculation, improve operation efficiency, can effectively judge operating bearing state.
Claims (1)
1. the bearing fault characteristics of Non-negative Matrix Factorization extracts and a diagnostic method, it is characterized in that, comprises the following steps:
1) utilize acceleration transducer to gather the vibration signal x of rolling bearing
i(n), n is sampling number, and n is greater than zero integer, and i is signal numbering;
2) utilize Short Time Fourier Transform, calculate vibration signal x
i(n) time-frequency distributions S
i, wherein slippage window is long for [1/2
3* n, 1/2*n];
3) to the time-frequency distributions S obtaining through conversion
idelivery, only retains amplitude information | S
i|, and by all amplitude spectrums | S
i| matrix-vector;
4), from i class signal, choose arbitrarily the amplitude spectrum composing training collection after a part of vectorization
V
train(i)={υec(|S
1|),υec(|S
2|),...,υec(|S
t|)} (1)
5) according to the definition of Non-negative Matrix Factorization
To V
train(i) decompose, obtain
V
train(i)≈W
n×r(i)H
r×m(i) (3)
W wherein
n * r(i) be basis matrix, be considered as the feature of extraction, and H
r * m(i) be weight coefficient matrix;
6) by the basis matrix W of various types of signal
n * r(i) assembled arrangement is the over-complete dictionary of atoms W of a rolling bearing fault
train,
W
train={W
n×r(1),W
n×r(2),...,W
n×r(T)} (6)
7) extract vibration-testing signal y (n), by Short Time Fourier Transform, obtain the time-frequency spectrum that only comprises amplitude information | Y|;
8) by amplitude spectrum | Y| is to former word bank W of training stage
trainprojection, according to the Clustering features of Non-negative Matrix Factorization, realizes Fault Identification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310693212.8A CN103728135A (en) | 2013-12-16 | 2013-12-16 | Bearing fault feature extraction and diagnosis method of non-negative matrix factorization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310693212.8A CN103728135A (en) | 2013-12-16 | 2013-12-16 | Bearing fault feature extraction and diagnosis method of non-negative matrix factorization |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103728135A true CN103728135A (en) | 2014-04-16 |
Family
ID=50452305
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310693212.8A Pending CN103728135A (en) | 2013-12-16 | 2013-12-16 | Bearing fault feature extraction and diagnosis method of non-negative matrix factorization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103728135A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109214469A (en) * | 2018-10-24 | 2019-01-15 | 西安交通大学 | A kind of source signal separation method based on non-negative tensor resolution |
CN109298633A (en) * | 2018-10-09 | 2019-02-01 | 郑州轻工业学院 | Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization |
CN110319995A (en) * | 2019-08-14 | 2019-10-11 | 清华大学 | Firer's shock response data time-frequency spectrum analysis method |
CN111189624A (en) * | 2020-01-08 | 2020-05-22 | 中国工程物理研究院总体工程研究所 | Method for identifying loosening state of bolt connection structure based on vibration signal time-frequency characteristics |
CN114942133A (en) * | 2022-05-20 | 2022-08-26 | 大连理工大学 | Optimal rank non-negative matrix factorization-based early fault diagnosis method for planetary gearbox |
CN117302236A (en) * | 2023-09-27 | 2023-12-29 | 湖北天凯风林电子有限公司 | Vehicle state monitoring method and system based on deep learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101853239A (en) * | 2010-05-06 | 2010-10-06 | 复旦大学 | Nonnegative matrix factorization-based dimensionality reducing method used for clustering |
US20120316886A1 (en) * | 2011-06-08 | 2012-12-13 | Ramin Pishehvar | Sparse coding using object exttraction |
CN102879196A (en) * | 2012-09-25 | 2013-01-16 | 西安交通大学 | Compound fault diagnosing method for planetary gearbox by using matrix wavelet transformation |
CN102998118A (en) * | 2012-11-29 | 2013-03-27 | 西安交通大学 | Bearing quantitative diagnosis method based on morphological filtering and complexity measure |
EP2581725A2 (en) * | 2011-10-13 | 2013-04-17 | General Electric Company | Methods and systems for automatic rolling-element bearing fault detection |
CN103366182A (en) * | 2013-07-05 | 2013-10-23 | 西安电子科技大学 | Face recognition method based on all-supervision non-negative matrix factorization |
-
2013
- 2013-12-16 CN CN201310693212.8A patent/CN103728135A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101853239A (en) * | 2010-05-06 | 2010-10-06 | 复旦大学 | Nonnegative matrix factorization-based dimensionality reducing method used for clustering |
US20120316886A1 (en) * | 2011-06-08 | 2012-12-13 | Ramin Pishehvar | Sparse coding using object exttraction |
EP2581725A2 (en) * | 2011-10-13 | 2013-04-17 | General Electric Company | Methods and systems for automatic rolling-element bearing fault detection |
CN102879196A (en) * | 2012-09-25 | 2013-01-16 | 西安交通大学 | Compound fault diagnosing method for planetary gearbox by using matrix wavelet transformation |
CN102998118A (en) * | 2012-11-29 | 2013-03-27 | 西安交通大学 | Bearing quantitative diagnosis method based on morphological filtering and complexity measure |
CN103366182A (en) * | 2013-07-05 | 2013-10-23 | 西安电子科技大学 | Face recognition method based on all-supervision non-negative matrix factorization |
Non-Patent Citations (3)
Title |
---|
施蓓琦 等: "应用稀疏非负矩阵分解聚类实现高光谱影像波段的优化选择", 《测绘学报》 * |
李兵等: "二维非负矩阵分解在齿轮故障诊断中的应用", 《振动、测试与诊断》 * |
栾美洁: "基于非负矩阵分解的机械故障信号特征分析的研究", 《万方数据》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109298633A (en) * | 2018-10-09 | 2019-02-01 | 郑州轻工业学院 | Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization |
CN109214469A (en) * | 2018-10-24 | 2019-01-15 | 西安交通大学 | A kind of source signal separation method based on non-negative tensor resolution |
CN110319995A (en) * | 2019-08-14 | 2019-10-11 | 清华大学 | Firer's shock response data time-frequency spectrum analysis method |
CN111189624A (en) * | 2020-01-08 | 2020-05-22 | 中国工程物理研究院总体工程研究所 | Method for identifying loosening state of bolt connection structure based on vibration signal time-frequency characteristics |
CN114942133A (en) * | 2022-05-20 | 2022-08-26 | 大连理工大学 | Optimal rank non-negative matrix factorization-based early fault diagnosis method for planetary gearbox |
CN114942133B (en) * | 2022-05-20 | 2023-04-14 | 大连理工大学 | Optimal rank non-negative matrix factorization-based early fault diagnosis method for planetary gearbox |
CN117302236A (en) * | 2023-09-27 | 2023-12-29 | 湖北天凯风林电子有限公司 | Vehicle state monitoring method and system based on deep learning |
CN117302236B (en) * | 2023-09-27 | 2024-03-26 | 湖北天凯风林电子有限公司 | Vehicle state monitoring method and system based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Han et al. | Multi-level wavelet packet fusion in dynamic ensemble convolutional neural network for fault diagnosis | |
CN105760839A (en) | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine | |
CN105275833B (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
CN103728135A (en) | Bearing fault feature extraction and diagnosis method of non-negative matrix factorization | |
CN103728551B (en) | A kind of analog-circuit fault diagnosis method based on cascade integrated classifier | |
CN110543860B (en) | Mechanical fault diagnosis method and system based on TJM (machine learning model) transfer learning | |
CN105115594B (en) | Gear-box vibration signal fault signature extracting method based on Wavelet Entropy and information fusion | |
CN106778694A (en) | A kind of gear transmission noises analysis method based on set empirical mode decomposition and SVMs | |
CN103245907B (en) | A kind of analog-circuit fault diagnosis method | |
CN112418013B (en) | Complex working condition bearing fault diagnosis method based on meta-learning under small sample | |
CN109977920A (en) | Fault Diagnosis of Hydro-generator Set method based on time-frequency spectrum and convolutional neural networks | |
CN104020401B (en) | The appraisal procedure of transformer insulated heat ageing state based on cloud models theory | |
CN104089774B (en) | A kind of gear failure diagnosing method based on the parallel orthogonal coupling of multiword allusion quotation | |
CN110705456A (en) | Micro motor abnormity detection method based on transfer learning | |
CN110298085A (en) | Analog-circuit fault diagnosis method based on XGBoost and random forests algorithm | |
CN105678343A (en) | Adaptive-weighted-group-sparse-representation-based diagnosis method for noise abnormity of hydroelectric generating set | |
CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
CN104504403B (en) | A kind of rotating machinery fault Forecasting Methodology based on scattering conversion | |
Liu et al. | Multi-feature fusion for fault diagnosis of rotating machinery based on convolutional neural network | |
CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
CN115221930A (en) | Fault diagnosis method for rolling bearing | |
CN112146882B (en) | Bearing fault diagnosis method based on transfer learning vibration signal image recognition | |
Jin et al. | Fault diagnosis of bearing based on refined piecewise composite multivariate multiscale fuzzy entropy | |
CN103592595A (en) | Fault diagnosis method of switching current circuit | |
Yu et al. | Rolling bearing fault feature extraction and diagnosis method based on MODWPT and DBN |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140416 |