CN103364024B - Based on the sensor fault diagnosis method of empirical mode decomposition - Google Patents
Based on the sensor fault diagnosis method of empirical mode decomposition Download PDFInfo
- Publication number
- CN103364024B CN103364024B CN201310296174.2A CN201310296174A CN103364024B CN 103364024 B CN103364024 B CN 103364024B CN 201310296174 A CN201310296174 A CN 201310296174A CN 103364024 B CN103364024 B CN 103364024B
- Authority
- CN
- China
- Prior art keywords
- variance
- fault
- imf
- surplus
- sensor
- 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
Links
Abstract
The invention belongs to industry monitoring field, especially the sensor fault diagnosis method based on empirical mode decomposition being applied to power plant's thermal parameter is applicable to, the present invention carries out empirical mode decomposition by the data in the time series measured by the sensor to monitoring thermal parameter, and variance calculating is carried out to each curve (intrinsic mode functions IMF and surplus) after decomposing, a feature be reflected in numerically can be obtained, if the process afterwards for these data does not meet this variation characteristic, just prove to there occurs fault.Algorithm of the present invention is simple, and algorithm itself, only for time series, thus can process the data of any unit, each type parameters such as such as temperature, pressure, flow, power.Calculation procedure is easy, can obtain diagnostic result fast, without the network of complex structure.Especially can obtain the variation tendency of any parameter on following certain hour, with other fault diagnosis algorithm conbined usage, also there is certain using value.<!--1-->
Description
Technical field
The invention belongs to industry monitoring field, be specially the application of empirical mode decomposition algorithm in sensor fault diagnosis direction, be especially applicable to the fault diagnosis being applied to power plant's thermal parameter.
Background technology
In the face of the power plant thermal system that unit capacity increases day by day, along with the development of Automated condtrol, whole power plant starts more and more " modernization ".But this also represents in whole Monitor System on Power Plant, the sensor of installation gets more and more.No matter be real-time Monitoring and Controlling, or energy consumption calculation is carried out in data acquisition, the collection of a series of thermal parameters such as power, temperature, pressure, flow is all on the Measurement accuracy of the sensor being based upon numerous " basic unit ".Once sensor experiences failure, the accuracy of follow-up real-time control and performance calculating directly will be had influence on.Sensor fault diagnosis method is the study hotspot in industry monitoring field always.
Empirical mode decomposition can extract one group of data variation tendency in time well, is suggested and is just widely used in a lot of field, achieve good effect since 1998.But not yet the method is applied to this field of thermal parameter fault diagnosis.
Summary of the invention
For overcoming the above problems, the invention provides a sensor fault diagnosis method based on empirical mode decomposition, the variation tendency of certain parameter in following certain hour can be extracted rapidly and accurately, thus judge whether sensor has fault, and it is any to judge to belong in " complete failure, precise decreasing, drifting fault, droop " these four kinds of sensor most common failures.
The present invention adopts following technical scheme:
Based on the sensor fault diagnosis method of empirical mode decomposition, comprise the steps,
(1) original signal of M the data point recorded in certain hour t for arbitrary sensor carries out empirical mode decomposition, obtains the curve of a n IMF and surplus,
Wherein
s (t) is original signal, x
it () is IMF, r
nt () is surplus;
(2) n the IMF decomposited step (1) and each curve of surplus carry out variance calculating respectively, and calculate the variance of original signal, the variance calculating each IMF and surplus respectively with the ratio of original signal variance, the variance of each IMF and surplus and the ratio of original signal variance are the variation characteristic of original signal;
(3) new signal of record m data point is continued, each IMF of signal of M+m data point obtain signal repetition step (1) and the step (2) of a front and back M+m data point and the ratio of the signal of the variance of surplus and M+m data point are the variation characteristic of new signal, if the variation characteristic of new signal is consistent with the variation characteristic of original signal, then do not break down; If different, then there occurs fault.
Described in described step (3), the variation characteristic of new signal is consistent with the variation characteristic of original signal, refers to that front four IMF are consistent with the ratio of signal data variance with the variance of surplus separately, m >=M in described step (3).
Further, the variation characteristic of described new signal and the variation characteristic of original signal is consistent refers to that the two rate of change is no more than 100%.
Further, M>100 in described step (1).
Further, when breaking down, if when the variance of the first two IMF decomposited is increased to non-fault more than hundred times, then the fault occurred is sensor complete failure.
Further, if during non-fault, the variance that the n-th IMF starts it is less than 10% compared to the ratio of original signal variance; After breaking down, the variance of the n-th IMF equals the order of magnitude of the variance of (n-1) individual IMF, then the fault occurred is sensor accuracy decline fault.
Further, if when the variance of original signal and surplus is increased to non-fault after breaking down more than 2 times, the curve of surplus is in rising trend, then the fault occurred is sensor drift fault.
Further, when breaking down, if the first two IMF variance decomposited is almost constant, the variance of original signal and surplus slightly increases, and surplus curve still presents fluctuation tendency, then the fault occurred is sensor droop fault.
Further, if multiple associated sensor breaks down, feature description is not sensor fault, but unit itself breaks down.
Specific explanations is:
Step one: original signal s (t) recorded in certain time interval T for arbitrary sensor carries out empirical mode decomposition.
This wherein, the time of image data is indefinite, but the data bulk of record is preferably more than 100 points, is assumed to M here.
This wherein, the particular flow sheet of empirical mode decomposition algorithm is shown in Fig. 2.In order original signal s (t) is decomposed into n intrinsic mode functions, (i.e. IMF is designated as x to the result of empirical mode decomposition
i(t)) and a surplus, be also called trend term, represent the variation tendency of original signal, be designated as r
n(t).Have in theory
Step 2: variance calculating is carried out to each curve (n IMF and surplus) that previous step decomposites.And calculate the variance of original signal, compare the variance of each IMF and surplus and the magnitude relationship of original signal variance.
This wherein, the variance size of IMF represents the whether strong of thermal parameter fluctuation, and the variance of IMF is larger with the ratio of original signal variance, illustrates that the original signal Shaoxing opera that fluctuates is strong.And surplus represents the variation tendency of thermal parameter, its variance is larger with the ratio of original signal, describes original signal stronger in the trend of increase in future (or reduction).
This wherein, for different original signals, the ratio (number percent) that their IMF and the variance of surplus account for original signal variance is different, but is basically identical in the time series of oneself.A reflecting time Sequence Trend feature can be obtained like this.This feature is exactly the variation tendencies of these data in the past in that time and following certain hour, if do not meet this variation characteristic for the process of these data afterwards, just illustrates that sensor there occurs fault.
Step 3: the variation characteristic having been obtained original signal by previous step, continues record data after this, remembers that the data amount check newly collected is m.To front and back altogether M+m data point repeat the work of first two steps, new variation characteristic will be obtained.If this feature and result before, the decomposition result namely during M point is consistent, then do not break down.If from result is different before, then probably there occurs fault.
This wherein, the new data amount check m gathered is close with original data amount check M, or larger, could faults feature preferably.Such as M=100, and m=10, the variation tendency of rear like this 10 points is not enough to the data sequence significantly affecting whole 110 points, beats unless there occurs very great data, otherwise be cannot therefrom detect out of order.
This wherein, during non-fault, the variation characteristic of so-called new signal is consistent with original signal variation characteristic, and refer to main IMF and the variance of surplus, with the ratio of the variance of signal own, twice calculating in front and back should be consistent.Rate of change before and after can thinking in practical application is no more than 100%.Here allow the rate of change of 100% be because fault time increase the degree 10 times more than even 1,100 times mostly reduced, different faults just shows as different IMF and occurs changing.
This wherein, result and original signal are not inconsistent explanation and there occurs fault, if judged to break down, then the variance size and the surplus curvilinear trend that compare IMF just can failure judgement types.Which kind of, as there occurs fault, to contrast with following four kinds of main sensor fault forms of expression.
Complete failure fault signature: the variance of the first two IMF decomposited is increased to more than hundred times during non-fault.
Precise decreasing fault signature: during for non-fault, if the variance that the n-th IMF starts it is less than 10% compared to original signal variance; After this fault occurs, the variance of the n-th IMF and the variance of (n-1) individual IMF have same order.This is because precise decreasing represents the enhancing of fluctuation.
Drifting fault feature: the variance of original signal and surplus is increased to more than 2 times during non-fault after breaking down, the curve of surplus is in rising trend simultaneously.
And droop is characterised in that: the first two IMF variance decomposited is almost constant, the variance of original signal and surplus slightly increases, and surplus curve still presents fluctuation tendency.
In fault type, complete failure refers to that, from certain time, sensing data is no longer with actual change, but straight line.Precise decreasing refers to that accuracy reduces, and namely the constant variance of average increases.During drifting fault, the gap of measured value and actual value increases in time.Droop is then the difference that between measured value and actual value, existence one is invariable.
The present invention carries out empirical mode decomposition to the data recorded in certain hour, and variance calculating is carried out to each curve (intrinsic mode functions IMF and surplus) after decomposing, a feature be reflected in numerically can be obtained, this feature is exactly the variation tendencies of these data in the past in that time and following certain hour, if the process afterwards for these data does not meet this variation characteristic, just prove to there occurs fault.Algorithm of the present invention is simple, and algorithm itself, only for time series, thus can process the data of any unit, each type parameters such as such as temperature, pressure, flow, power.Calculation procedure is easy, can obtain diagnostic result fast, without the network of complex structure.Especially can obtain the variation tendency of any parameter on following certain hour, with other fault diagnosis algorithm conbined usage, also there is certain using value.
Accompanying drawing explanation
Fig. 1 is the method for diagnosing faults process flow diagram based on empirical mode decomposition.
Fig. 2 is the process flow diagram of empirical mode decomposition algorithm.
Fig. 3-Fig. 8 is that in the embodiment of the present invention, main steam pressure 100 some decomposition result curve maps gather (ordinate unit is MPa, and horizontal ordinate is number a little), and Fig. 3-Fig. 8 is followed successively by original signal, surplus, IMF1, IMF2, IMF3, IMF4.
Fig. 9-Figure 14 is that in the embodiment of the present invention, main steam pressure 140 some decomposition result curve maps gather (ordinate unit is MPa, and horizontal ordinate is number a little), and Fig. 9-Figure 14 is followed successively by original signal, surplus, IMF1, IMF2, IMF3, IMF4.
Figure 15-Figure 20 is that in the embodiment of the present invention, main steam pressure 200 some decomposition result curve maps gather (ordinate unit is MPa, and horizontal ordinate is number a little), and Figure 15-Figure 20 is followed successively by original signal, surplus, IMF1, IMF2, IMF3, IMF4.
Embodiment
As shown in Figure 1, diagnostic method proposed by the invention is mainly divided into two large modules.
First module calculates the original signal data of sensor record; Second module adds that the new signal data sequence that the new partial data gathered forms carries out same calculating to original signal.
Key is that if do not have fault to occur, this result of twice should be consistent, even if thermal parameter itself is in unordered change.And once occur inconsistent, concrete fault type can be judged according to different sensor fault features.
Below using the decomposition computation result of the main steam pressure force data of certain Power Plant as explanation.
Step (1): the calculating data collected being carried out to empirical mode decomposition and variance, as table 1 (main steam pressure for certain 1000MW unit).Here respectively 100,140 and 200 points are processed, the variance ratio can finding out their IMF and surplus and original signals is separately much the same, and namely the variance of IMF mentioned above and surplus and the variance ratio rate of change of signal itself are about 100%.Which ensure that this thermal parameter should develop according to this rule within this period of time, if it is trouble-free for also illustrating that data develop down after this manner.
Fig. 3-Figure 20 gathers the empirical mode decomposition result curve figure of 100,140 and 200 these three groups of data of point.In figure, ordinate is MPa, and horizontal ordinate is number a little.Table 1 is the summary sheet of variance result of calculation.
This wherein, empirical mode decomposition may decomposite 5 to 6 IMF, or more, but except first 2,3, the variance of the IMF after ranking is comparatively too little compared with original signal, generally less than 10%, so do not possess break-up value in the fault diagnosis carried in this method.In other words, even if the fault of there occurs, fault signature significantly appears in the variance change of front 2 IMF and surplus, and IMF variance afterwards may not possess increase very intuitively or reduce, and thus in the curve map decomposited, is only enumerated to IMF4 and just finishes.In variance Macro or mass analysis table, represent these IMF with " default " item.
The variation characteristic of certain 1000MW unit main steam pressure of table 1
The second step of this method carries out check processing in real time, if the data newly obtained have the variation tendency identical with upper table after processing, just illustrates that sensor is normal work.Otherwise, if the fault of there occurs, different features will be shown, be mainly these four kinds of typical sensors fault phase characteristics of correspondence of complete failure, precise decreasing, drifting fault and droop.Such as in this instance, to the comparative result after each typical fault process, as table 2-table 5:
The analysis of variance table of table 2 main steam pressure complete failure
| Original signal | IMF1 | IMF2 | Default | Trend | |
| Non-fault | 4.3018 | 0.2555 | 0.2379 | 4.1896 | |
| 30% place's fault | 1.4710 | 4.8344*10 5 | 5.543*10 5 | 23.0697 |
As table 2, during complete failure, IMF1 and IMF2 significantly increases, by during non-fault less than 1, become 10
5the order of magnitude, fault is fairly obvious.
The analysis of variance table of table 3 main steam pressure precise decreasing
| Original signal | IMF1 | IMF2 | IMF3 | Default | Trend | |
| Non-fault | 4.3018 | 0.2555 | 0.2379 | 0.0354 | 4.1896 | |
| 30% place's fault | 6.3157 | 1.2579 | 0.9971 | 0.3783 | 3.9497 |
As table 3, during non-fault, IMF1 and IMF2 is originally as about 0.2, becomes about 1 when breaking down, IMF3 more originally only have 1% of original signal variance less than, be do not possess break-up value, increase now 10 times, precise decreasing causes the enhancing of fluctuation just.
The analysis of variance table of table 4 main steam pressure drifting fault
| Original signal | IMF1 | IMF2 | Default | Trend | |
| Non-fault | 4.3018 | 0.2555 | 0.2379 | 4.1896 | |
| 30% place's fault | 18.7585 | 0.2661 | 0.2515 | 18.7699 |
As table 4, when front 2 IMF are constant, original signal turns over several times with trend term with same huge amplitude and increases, and surplus curve is again ascendant trend, meets drifting fault feature.
The analysis of variance table of table 5 main steam pressure droop
| Original signal | IMF1 | IMF2 | Default | Trend | |
| Non-fault | 4.3018 | 0.2555 | 0.2379 | 4.1896 | |
| 30% place's fault | 8.3736 | 0.2633 | 0.3029 | 8.3514 |
As table 5, IMF is almost constant, and original signal and trend term increase, but amplitude is not as drifting fault, more importantly the oscillogram of surplus is not latter half of steady, but there is fluctuation, ascendant trend neither terminate, be thus judged as droop instead of drifting fault.
Claims (7)
1. based on the sensor fault diagnosis method of empirical mode decomposition, it is characterized in that: comprise the steps,
(1) original signal of M the data point recorded in certain hour t for arbitrary sensor carries out empirical mode decomposition, obtains the curve of a n IMF and surplus,
Wherein
s (t) is original signal, x
it () is IMF, r
nt () is surplus;
(2) n the IMF decomposited step (1) and each curve of surplus carry out variance calculating respectively, and calculate the variance of original signal, the variance calculating each IMF and surplus respectively with the ratio of original signal variance, the variance of each IMF and surplus and the ratio of original signal variance are the variation characteristic of original signal;
(3) new signal of record m data point is continued, each IMF of signal of M+m data point obtain signal repetition step (1) and the step (2) of a front and back M+m data point and the ratio of the signal of the variance of surplus and M+m data point are the variation characteristic of new signal, if the variation characteristic of new signal is consistent with the variation characteristic of original signal, then do not break down; If different, then there occurs fault;
Wherein, described in described step (3), the variation characteristic of new signal is consistent with the variation characteristic of original signal, refers to that front four IMF are consistent with the ratio of signal data variance with the variance of surplus separately, m >=M in described step (3).
2. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, is characterized in that: the variation characteristic of described new signal is consistent with the variation characteristic of original signal refers to that the two rate of change is no more than 100%.
3. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, it is characterized in that: when breaking down, if the variance of the first two IMF decomposited is increased to more than hundred times during non-fault, then the fault occurred is sensor complete failure.
4. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, is characterized in that: if during non-fault, and the variance that the n-th IMF starts it is less than 10% compared to the ratio of original signal variance; After breaking down, the variance of the n-th IMF equals the order of magnitude of the variance of (n-1) individual IMF, then the fault occurred is sensor accuracy decline fault.
5. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, it is characterized in that: if when the variance of original signal and surplus is increased to non-fault after breaking down more than 2 times, the curve of surplus is in rising trend, then the fault occurred is sensor drift fault.
6. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, it is characterized in that: when breaking down, if the first two IMF variance decomposited is almost constant, the variance of original signal and surplus slightly increases, surplus curve still presents fluctuation tendency, then the fault occurred is sensor droop fault.
7. the sensor fault diagnosis method based on empirical mode decomposition according to claim 1, is characterized in that: if multiple associated sensor breaks down, feature description is not sensor fault, but unit itself breaks down.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310296174.2A CN103364024B (en) | 2013-07-12 | 2013-07-12 | Based on the sensor fault diagnosis method of empirical mode decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310296174.2A CN103364024B (en) | 2013-07-12 | 2013-07-12 | Based on the sensor fault diagnosis method of empirical mode decomposition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103364024A CN103364024A (en) | 2013-10-23 |
| CN103364024B true CN103364024B (en) | 2015-10-28 |
Family
ID=49365891
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310296174.2A Active CN103364024B (en) | 2013-07-12 | 2013-07-12 | Based on the sensor fault diagnosis method of empirical mode decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103364024B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104390657B (en) * | 2014-11-05 | 2017-12-12 | 浙江大学 | A kind of Generator Unit Operating Parameters measurement sensor fault diagnosis method and system |
| CN106021545B (en) * | 2016-05-27 | 2019-12-24 | 大连楼兰科技股份有限公司 | Method for vehicle remote diagnosis and spare part retrieval |
| CN109241915B (en) * | 2018-09-11 | 2022-03-25 | 浙江大学 | Intelligent power plant pump fault diagnosis method based on vibration signal stability and non-stationarity judgment and feature discrimination |
| CN109582913B (en) * | 2018-11-20 | 2023-08-22 | 国家海洋局第一海洋研究所 | Method for empirical mode decomposition screening iteration process termination criterion |
| CN116399402B (en) * | 2023-04-18 | 2024-01-23 | 南京晓庄学院 | Fault early warning system of wireless sensor for ecological environment monitoring |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859146A (en) * | 2010-07-16 | 2010-10-13 | 哈尔滨工业大学 | Satellite fault prediction method based on predictive filtering and empirical mode decomposition |
| CN102944416A (en) * | 2012-12-06 | 2013-02-27 | 南京匹瑞电气科技有限公司 | Multi-sensor signal fusion technology-based fault diagnosis method for wind turbine blades |
| CN103200139A (en) * | 2013-04-11 | 2013-07-10 | 西安电子科技大学 | OFDM signal bandwidth blind estimating method |
-
2013
- 2013-07-12 CN CN201310296174.2A patent/CN103364024B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101859146A (en) * | 2010-07-16 | 2010-10-13 | 哈尔滨工业大学 | Satellite fault prediction method based on predictive filtering and empirical mode decomposition |
| CN102944416A (en) * | 2012-12-06 | 2013-02-27 | 南京匹瑞电气科技有限公司 | Multi-sensor signal fusion technology-based fault diagnosis method for wind turbine blades |
| CN103200139A (en) * | 2013-04-11 | 2013-07-10 | 西安电子科技大学 | OFDM signal bandwidth blind estimating method |
Non-Patent Citations (2)
| Title |
|---|
| A novel denoising method and its application to MEMS gyro signal;程向红等;《中国控制与决策会议论文》;20130527;正文第3248-3252页及摘要 * |
| Fault Diagnosis for Satellite’s Attitude Determination System Based on Model Error Prediction and EMD;wang jiongqi等;《2013美国控制会议》;20130619;正文第4337-4342页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103364024A (en) | 2013-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103364024B (en) | Based on the sensor fault diagnosis method of empirical mode decomposition | |
| CN109446187B (en) | Method for monitoring health state of complex equipment based on attention mechanism and neural network | |
| CN103064340B (en) | Failure prediction method facing to numerically-controlled machine tool | |
| Chen et al. | Observer-based FDI schemes for wind turbine benchmark | |
| CN101907088B (en) | Fault diagnosis method based on one-class support vector machines | |
| CN103776480A (en) | Small-fault detection method and device based on multiple moving average | |
| CN102999038B (en) | The diagnostic device of generating set and the diagnostic method of generating set | |
| CN102426862B (en) | Modeling method for reliability of reactor-shutdown protecting signal in DCS (digitizer control system) and system | |
| CN111506049B (en) | Multiple fault diagnosis method for aero-engine control system based on AANN network system | |
| CN104390657A (en) | Generator set operating parameter measuring sensor fault diagnosis method and system | |
| CN106250709A (en) | Gas turbine abnormality detection based on sensors association network and fault diagnosis algorithm | |
| Yan et al. | Fault diagnosis of rotating machinery equipped with multiple sensors using space-time fragments | |
| CN106226074A (en) | Based on convolutional neural networks and the rotary machinery fault diagnosis method of small echo gray-scale map | |
| CN111946559B (en) | Method for detecting structures of wind turbine foundation and tower | |
| CN111474919A (en) | Aeroengine control system sensor fault diagnosis method based on AANN network group | |
| CN202648860U (en) | Pressure transmitter of process fluid pressure during measuring industry process and process variable transmitter | |
| CN102222151A (en) | Analog circuit fault prediction method based on ARMA (Autoregressive Moving Average) | |
| CN106067032A (en) | Evaluation methodology for the sensor system of selection of data exception monitoring | |
| CN104008294A (en) | System and method for detecting abnormality of bearing | |
| CN104182623A (en) | Thermal process data detection method based on equivalent change rate calculation | |
| CN103794257A (en) | Nuclear power plant first-out alarm processing method and system | |
| CN104615123B (en) | K-nearest neighbor based sensor fault isolation method | |
| CN103530660B (en) | A kind of strip tension sensor fault method of early diagnosis | |
| CN107222169A (en) | A kind of photovoltaic array based on outlier detection is monitored and fault detection method in real time | |
| Zhong et al. | An improved correlation-based anomaly detection approach for condition monitoring data of industrial equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |