CN109145773B - Fault prediction method for multi-source trend information fusion - Google Patents

Fault prediction method for multi-source trend information fusion Download PDF

Info

Publication number
CN109145773B
CN109145773B CN201810866101.5A CN201810866101A CN109145773B CN 109145773 B CN109145773 B CN 109145773B CN 201810866101 A CN201810866101 A CN 201810866101A CN 109145773 B CN109145773 B CN 109145773B
Authority
CN
China
Prior art keywords
fault
trend
growth
failure
phi
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810866101.5A
Other languages
Chinese (zh)
Other versions
CN109145773A (en
Inventor
谭晓栋
左明健
黄娟
周梓鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201810866101.5A priority Critical patent/CN109145773B/en
Publication of CN109145773A publication Critical patent/CN109145773A/en
Application granted granted Critical
Publication of CN109145773B publication Critical patent/CN109145773B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Abstract

The invention discloses a fault prediction method for multi-source trend information fusion, which comprises the following steps: s1, acquiring multi-source fault growth data; s2, preprocessing the original data; s3, determining a fault prediction time node; s4, constructing a fault growth trend; s5, fusing fault growth trend information; and S6, predicting the fault growth trend. According to the method, the measuring points which can effectively track the fault growth process can be selected by analyzing the fault growth trend characteristics described by different measuring points, the influence of measuring point data with insignificant trend characteristics and large random interference on the fault prediction result is reduced, and the fault prediction accuracy is improved; by analyzing the correlation degree of the growth trend of each fault, the influence of redundant measurement points with similar trend characteristics on the fault prediction efficiency can be eliminated, and the prediction time is reduced; by fusing the fault growth trend described by a plurality of effective measuring points, the limitation of relying on the trend information of a single measuring point is avoided, and the precision of fault prediction is improved.

Description

Fault prediction method for multi-source trend information fusion
Technical Field
The invention belongs to the field of fault detection, and particularly relates to a fault prediction method for multi-source trend information fusion. A novel fault prediction method is mainly provided for typical mechanical parts (such as bearings, gears, shafts and the like) in an electromechanical system.
Background
The equipment failure prediction technology can not only avoid economic loss caused by misappropriation of the conventional planned maintenance strategy, but also effectively reduce failures, avoid waste, ensure continuous and efficient operation of key equipment in industrial production and improve the productivity. For fault prediction, monitoring and tracking of a fault growth trend are important to ensure prediction accuracy. How to obtain as much and as accurately as possible information about the trend of the increase of the fault from the original signal is the basis for the fault prediction. Any fault type is incomplete by using one-side data information to reflect the trend behavior. In order to make more accurate and comprehensive fault prediction, more fault growth trend information needs to be acquired from different angles, but as the information quantity increases, the problem which needs to be solved urgently is how to acquire and fuse the fault growth trend information sufficiently to improve the accuracy of fault prediction.
At present, the known fault prediction method focuses on data processing, feature extraction and prediction methods, and adopts a time domain, a frequency domain or a time-frequency domain combined method to extract fault features from an original signal interfered by complex noise, and then adopts an intelligent algorithm to predict the time of occurrence of a fault. However, the known method has the following problems:
firstly, how to acquire the largest fault growth trend information associated with prediction is not comprehensively considered from the source of prediction data acquisition in the conventional fault prediction technology, so that a large amount of information irrelevant to fault prediction is used as a fault prediction data source, and the support degree of the data source to the fault prediction technology is weak;
secondly, due to different measuring point positions, the acquired data sources have different fault growth trend information degrees or the data sources have different degrees of interference from environmental noise, even if the same feature extraction method is adopted, the fault growth trends are greatly different, and some measuring points are more obvious to the whole process of the fault growth trend but are not sensitive to early faults; some of the stations are sensitive to early faults, but the overall fault growth trend described fluctuates significantly. The existing fault prediction technology cannot integrate the fault growth trend characteristics described by multi-source measuring points, so that the fault prediction precision is influenced.
Disclosure of Invention
The invention aims to overcome the defects that the accuracy and precision of fault prediction are low and the like caused by the fact that the conventional fault prediction method lacks fault growth characteristics which are obtained from a data source and described by fusing different measuring points, and the invention aims to provide a fault prediction method with multi-source trend information fusion. The method comprises the steps of establishing fault growth trends described by different measuring points, analyzing different trend characteristics, selecting the fault growth trend with remarkable trend characteristics as a data source for fault prediction, fusing the fault growth trends described by a plurality of effective measuring points according to trend capacity to serve as historical data of a fault prediction model, and predicting the time of occurrence of a fault by adopting a time sequence prediction method. The invention finally provides an effective strategy for tracking and monitoring the fault growth trend, and provides an effective data source and fault growth trend fusion characteristic for the fault prediction method, thereby ensuring that the fault prediction precision is more efficient and accurate.
The technical scheme adopted by the invention is as follows:
a multi-source trend information fusion fault prediction method comprises the following steps:
s1, multi-source fault growth data acquisition: collecting raw data of a fault evolution process at the same time interval by using a plurality of available measuring points;
s2, preprocessing the original data, and removing a trend item and an interference item;
s3, determining a fault prediction time node;
s4, constructing a fault growth trend: calculating a fault trend characteristic value of the original data at each moment, establishing a fault growth trend described by all measuring points, analyzing trend characteristics of the fault growth trend, and selecting a fault growth trend set with obvious monotone increasing; calculating the correlation degree of the fault growth trend in the set, and selecting a fault growth trend set without significant correlation;
s5, fusing fault growth trend information: fusing no significant related fault growth trend in the step S4 to obtain a redundancy-free fault growth trend set;
s6, using the redundancy-free fault growth trend set obtained in the step S5 as the input of the fault prediction model, and performing the following fault growth trend prediction:
a. predicting a fault trend characteristic value after the fault prediction time node;
b. and determining the time of the failure prediction time node to complete failure.
Specifically, in the step S1, raw data of a fault growth process is collected by sensors disposed at each measurement point, where the fault growth process is a whole fault growth process from a normal state to a complete failure state, and the specific implementation method is as follows:
preliminary deployment measurement point
Figure GDA0003187991260000031
NTCollecting the life-span data of the system from normal state to complete failure state for the total number of the measuring points
Figure GDA0003187991260000032
Wherein, OiFor the ith measuring point tiCollected fault evolution data.
Further, the step S2 is to preprocess the raw data, and the specific implementation method for removing the trend term and the interference term is as follows:
Figure GDA0003187991260000033
in the formula, Xi(n) is data OiData at time n,. mu.iAnd σiAre each XiMean and standard deviation of (n).
Further, the specific implementation method of determining the failure prediction time node in step S3 is as follows:
s31, calculating a fault early warning characteristic value of the original data at the time t, and calculating the fault early warning characteristic value of the data from the time 0-t by using a formula (2) to obtain a fault early warning characteristic value of the data O:
FP(t)＝fFP(O(t)) (2)
in the formula, FP (t) is a fault early warning characteristic value corresponding to the time t, O (t) is data collected at the time t, fFP() computing a function for the fault warning features;
s32, judging the fault early warning at the current moment, and calculating a fault prediction time node according to the formula (3):
Figure GDA0003187991260000041
in the formula (f)FP(O (t)) is a fault early warning characteristic value corresponding to the time t, FFPFor early warning threshold in early failure state, TTP is fFP(O (t)) and FFPAnd the relative error is less than or equal to 5 percent.
Preferably, the fault early warning feature calculation function is a time domain index, a frequency domain index or a time-frequency domain index;
the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, skewness, kurtosis, variance, a peak value, a standard deviation, a peak-to-peak value, average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index;
the frequency domain index is a power spectrum or a high-order statistic;
the time-frequency domain index is wavelet entropy or energy entropy.
Further, the specific implementation method of step S4 is as follows:
s41, calculating the fault trend characteristic value of the original data at each moment by using the formula (4):
FT(t)＝fFT(O(t)) (4)
in the formula, FT (t) is a fault trend characteristic value corresponding to t time, O (t) is data collected at t time, fFT() computing a function for the fault trend signature;
s42, smoothing FT (t), and establishing the fault growth trend described by all measuring points
Figure GDA0003187991260000042
NTAs a total number of measured points, [ phi ]i＝{FT(1),FT(2),...,FT(NL) The ith measuring point is used for establishing a fault growth trend;
s43, analyzing trend characteristics of the fault growth trend, removing measuring points with no trend characteristics and decreasing fault growth trend, selecting the fault growth trend with obvious monotone increasing, and calculating the fault growth trend phi according to formula (5)iM-K statistic Zi：
Figure GDA0003187991260000051
In the formula, ZiTo a failure growth tendency phiiM-K statistic of (N)LIs phiiNumber of samples of fault-evolution data collected in the process, SiTo a failure growth tendency phiiTest statistics, calculated according to equation (6):
Figure GDA0003187991260000052
wherein sign is a sign function when FT (n)1)-FT(n2) Sign (FT (n) when it is less than, equal to, or greater than zero1)-FT(n2) Are each-1, 0 or 1, NLIs OiThe total number of collected fault evolution data samples;
removing the trend-free features and the decreasing fault growth trend according to the formula (7), and keeping the fault growth trend set phi with the characteristic of obvious increasing trend with 95 percent of confidencets：
Figure GDA0003187991260000053
In the formula phiiThe increasing trend of the fault described for the ith station, ZiTo a failure growth tendency phiiM-K statistic of (N)TThe total number of the measuring points is;
s44, calculating phi according to the formula (8)tsMiddle tendency of phiiAnd phijDegree of correlation rijSelecting a set of fault growth trends without significant correlation:
Figure GDA0003187991260000054
in the formula phitsFor a set of failure growth trends with a significantly increasing trend characteristic, ΦiIs phitsThe increasing trend of the fault described by the ith station, i.e. phii∈Φts，NLIs OiThe total number of the collected fault growth data samples is removedtsMiddle | rijThe failure growth trend with the significant correlation of the value greater than 0.8 is obtained, and a failure growth trend set phi without redundancy trend characteristics is obtainedrs。
Preferably, the fault tendency characteristic calculation function is a time domain index, a frequency domain index or a time-frequency domain index;
the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, skewness, kurtosis, variance, a peak value, a standard deviation, a peak-to-peak value, average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index;
the frequency domain index is a power spectrum or a high-order statistic;
the time-frequency domain index is wavelet entropy or energy entropy.
Further, the specific implementation method of step S5 is as follows:
s51, calculating a failure growth trend set phi according to the formula (9)rsWeight ω of each trend ini：
Figure GDA0003187991260000061
In the formula, ZiSet phi of growth trends for redundancy-free faultsrsMiddle ith tendency ΦiM-K statistic of (N)rsIs phirsMedian failure growth trend total;
s52, using equation (10) for phirsAnd carrying out information fusion on the medium fault growth trend to obtain a fused fault growth trend:
Figure GDA0003187991260000062
in the formula phirsFor a set of no-redundancy failure growth trends, phiRFor failure growth trend after information fusion, omegaiTo a failure growth tendency phiiWeight of (1), NrsIs phirsTotal number of median failure growth trends, Φi∈Φrs；
S53, increasing trend phi of fused faultsRSmoothing is carried out to obtain the smooth fault growth trend phi*。
Further, the specific implementation method of step S6 is as follows:
s61 in phi*(1),...,Φ*(TTP) data is historical data of a fault growth trend prediction, the fault growth trend is used as an input of a prediction model, and TTP is an early fault prediction point;
s62, predicting the fault growth trend at the T moment after the TTP of the early fault prediction point by using a time series prediction method
Figure GDA0003187991260000071
S63, calculating the time from the early failure prediction point TTP to complete failure using equation (11):
Figure GDA0003187991260000072
wherein TTP is an early failure prediction point, FFTFor a failure characteristic threshold at full failure,
Figure GDA0003187991260000073
the TTF is the time from early failure prediction to TTP to complete failure, namely the predicted value of the failure prediction point TTP after the time T
Figure GDA0003187991260000074
And a failure threshold value FFTThe relative ratio of (A) is less than or equal to 0.05.
Preferably, the time series prediction method is exponential smooth prediction or exponential function fitting.
The invention has the beneficial effects that:
according to the method, the measuring points which can effectively track the fault growth process can be selected by analyzing the fault growth trend characteristics described by different measuring points, the influence of measuring point data with insignificant trend characteristics and large random interference on the fault prediction result is reduced, and the fault prediction accuracy is improved; by analyzing the correlation degree of the growth trend of each fault, the influence of redundant measurement points with similar trend characteristics on the fault prediction efficiency can be eliminated, and the prediction time is reduced; by fusing the fault growth trend described by a plurality of effective measuring points, measuring point information which can sensitively sense early faults and stably track the fault growth process can be integrated to be used as the input of a fault prediction model, the limitation of relying on single measuring point trend information is avoided, and the fault prediction precision is improved.
Drawings
FIG. 1 is a flow chart of a method of an embodiment of the present invention.
Fig. 2 is a roller bearing failure test stand according to an embodiment of the present invention.
Fig. 3 is a raw vibration signal collected at n-1 (i.e., 10 minutes of operation) at four stations in accordance with an embodiment of the present invention.
Fig. 4 shows the vibration signals of four measuring points of the example of the present invention after pretreatment with n equal to 1 (i.e. 10 minutes of operation).
FIG. 5 shows the corresponding failure early warning feature values of four measuring points according to the embodiment of the invention.
FIG. 6 shows the trend of the four points after smoothing the increase in the failure described by the embodiment of the present invention.
FIG. 7 is a graph of the failure growth trend curve phi after information fusion according to an embodiment of the present invention*。
FIG. 8 is a quadratic exponential smoothing prediction curve according to an embodiment of the present invention.
Detailed Description
The invention is further described with reference to the following figures and specific embodiments.
Example 1:
as shown in fig. 1, a failure prediction method for multi-source trend information fusion includes the following steps:
firstly, multi-source fault growth data acquisition, wherein original data of a fault evolution process are acquired at the same time interval by using a plurality of available measuring points.
In this embodiment, the original data of the fault evolution process is collected by the sensors deployed at each measuring point, the fault evolution process is the whole fault growth process from a normal state to a complete failure state, and the specific implementation method is as follows:
preliminary deployment measurement point
Figure GDA0003187991260000081
NTCollecting the life-span data of the system from normal state to complete failure state for the total number of the measuring points
Figure GDA0003187991260000091
Wherein, OiFor the ith measuring point tiCollected fault evolution data.
And secondly, preprocessing the original data to remove a trend item and an interference item.
Further, the step S2 is to preprocess the raw data, and the specific implementation method for removing the trend term and the interference term is as follows:
Figure GDA0003187991260000092
in the formula, Xi(n) is data OiData at time n,. mu.iAnd σiAre each XiMean and standard deviation of (n).
And thirdly, determining a fault prediction time node.
The specific implementation method comprises the following steps:
s31, calculating a fault early warning characteristic value of the original data at the time t, and calculating the fault early warning characteristic value of the data from the time 0-t by using a formula (2) to obtain a fault early warning characteristic value of the data O:
FP(t)＝fFP(O(t)) (2)
in the formula, FP (t) is a fault early warning characteristic value corresponding to the time t, O (t) is data collected at the time t, fFP() computing a function for the fault warning features;
s32, judging the fault early warning at the current moment, and calculating a fault prediction time node according to the formula (3):
Figure GDA0003187991260000093
in the formula (f)FP(O (t)) is a fault early warning characteristic value corresponding to the time t, FFPFor early warning in fault stateThreshold value, TTP is fFP(O (t)) and FFPAnd the relative error is less than or equal to 5 percent.
In this embodiment, the fault early warning feature calculation function is a time domain index, a frequency domain index or a time-frequency domain index; the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, a skewness, a kurtosis, a variance, a peak value, a standard deviation, a peak-peak value, an average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index; a frequency domain index power spectrum or a high order statistic; the time-frequency domain indexes are wavelet entropy and energy entropy.
Step four, constructing a fault growth trend: calculating a fault trend characteristic value of the original data at each moment, establishing a fault growth trend described by all measuring points, analyzing trend characteristics of the fault growth trend, and selecting a fault growth trend set with obvious monotone increasing; and calculating the correlation degree of the fault growth trend in the set, and selecting the fault growth trend set without significant correlation.
The specific implementation method comprises the following steps:
s41, calculating the fault trend characteristic value of the original data at each moment by using the formula (4):
FT(t)＝fFT(O(t)) (4)
in the formula, FT (t) is a fault trend characteristic value corresponding to t time, O (t) is data collected at t time, fFT() computing a function for the fault trend signature;
s42, smoothing FT (t), and establishing the fault growth trend described by all measuring points
Figure GDA0003187991260000101
NTAs a total number of measured points, [ phi ]i＝{FT(1),FT(2),...,FT(NL) The ith measuring point is used for establishing a fault growth trend;
s43, analyzing trend characteristics of the fault growth trend, removing measuring points with no trend characteristics and decreasing fault growth trend, selecting the fault growth trend with obvious monotone increasing, and calculating the fault growth trend phi according to formula (5)iM-K statistic Zi：
Figure GDA0003187991260000102
In the formula, ZiTo a failure growth tendency phiiM-K statistic of (N)LIs phiiNumber of samples of fault-evolution data collected in the process, SiTo a failure growth tendency phiiTest statistics, calculated according to equation (6):
Figure GDA0003187991260000103
wherein sign is a sign function when FT (n)1)-FT(n2) Sign (FT (n) when it is less than, equal to, or greater than zero1)-FT(n2) Are each-1, 0 or 1, NLIs OiThe total number of collected fault evolution data samples;
removing the trend-free features and the decreasing trend of the fault growth according to the formula (7), and preserving the fault growth trend set phi of the characteristic of obvious increasing trend with 95 percent of confidencets：
Figure GDA0003187991260000111
In the formula phiiThe increasing trend of the fault described for the ith station, ZiTo a failure growth tendency phiiM-K statistic of (N)TThe total number of the measuring points is;
s44, calculating phi according to the formula (8)tsMiddle tendency of phiiAnd phijDegree of correlation rijSelecting a set of fault growth trends without significant correlation:
Figure GDA0003187991260000112
in the formula phiiIs phitsThe increasing trend of the fault described by the ith station, i.e. phii∈Φts，NLIs OiThe total number of the collected fault growth data samples is removedtsMiddle | rijThe failure growth trend with the significant correlation of the value greater than 0.8 is obtained, and a failure growth trend set phi without redundancy trend characteristics is obtainedrs。
In this embodiment, the function for calculating the fault tendency characteristics is that the time domain index is a time domain index, a frequency domain index or a time-frequency domain index; the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, a skewness, a kurtosis, a variance, a peak value, a standard deviation, a peak-peak value, an average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index; the frequency domain index is a power spectrum or high-order statistic; the time-frequency domain index is wavelet entropy or energy entropy.
And fifthly, fusing fault growth trend information: and fusing the fault growth trends in the fault growth trend set which is not obviously related in the fourth step to obtain a fault growth trend set.
The specific implementation method comprises the following steps:
s51, calculating a failure growth trend set phi according to the formula (9)rsWeight ω of each trend ini：
Figure GDA0003187991260000121
In the formula, ZiSet of fault growth trends ΦrsMiddle ith tendency ΦiM-K statistic of (N)rsIs phirsMedian failure growth trend total;
s52, using equation (10) for phirsAnd carrying out information fusion on the medium fault growth trend to obtain a fused fault growth trend sequence:
Figure GDA0003187991260000122
in the formula phirsFor a set of fault growth trends after information fusion, omegaiTo a failure growth tendency phiiWeight of (1), NrsIs phirsMean time to failure growth trendTotal number,. phii∈Φrs；
S53, merging the fault growth trend sequences phiRSmoothing is carried out to obtain the smooth fault growth trend phi*。
And sixthly, taking the fault growth trend set obtained in the fifth step as the input of a fault prediction model, and predicting the following fault growth trends:
a. predicting a fault trend characteristic value after the fault prediction time node;
b. and determining the time of the failure prediction time node to complete failure.
The specific implementation method comprises the following steps:
s61 in phi*(1),...,Φ*(TTP) data is historical data of failure growth trend prediction, with failure growth trend as input to the prediction model;
s62, predicting the fault growth trend at the T moment after the TTP of the early fault prediction point by using a time series prediction method
Figure GDA0003187991260000123
S63, calculating the time from the early failure prediction point TTP to complete failure using equation (11):
Figure GDA0003187991260000131
wherein TTP is an early failure prediction point, FFTFor a failure characteristic threshold at full failure,
Figure GDA0003187991260000132
the TTF is the time from early failure prediction to TTP to complete failure, namely the predicted value of the failure prediction point TTP after the time T
Figure GDA0003187991260000133
And a failure threshold value FFTThe relative ratio of (A) is less than or equal to 0.05.
In this embodiment, the time series prediction method is exponential smoothing prediction, exponential function fitting, or least square method.
The main idea of the invention is explained in detail below using a roller bearing as an example:
firstly, multi-source fault growth data acquisition.
And acquiring a residual service life prediction data source by adopting a roller bearing fault test bench. Roller bearing failure test station as shown in fig. 2, in the figure, 1-first bearing; 2-a second bearing; 3-a third bearing; 4-a fourth bearing; 5-an accelerometer; 6-radial load; 7-thermocouple. Four bearings are arranged on the same shaft, a motor is driven to drive a driving shaft and the bearings to rotate, 1 measuring point is respectively arranged in the Y direction from left to right 4 bearing seats, and the 4 measuring points are defined as TI＝{t1,t2,t3,t4I.e. NTAnd 4, collecting vibration data of the bearing in the whole life process. The specific test process is as follows: in the process from the beginning of the test to the acquisition of the automatic shutdown data, vibration signals of the bearing in the Y direction are acquired through 4 measuring points at intervals of 10 minutes, the sampling frequency is 20KHz, 4096 points are acquired each time, and 984 sample numbers, namely the sample number N corresponding to the fault growth data, are acquiredLThe run time was 96200 minutes 984, and the collected data was collected by LabVIEW software. Four measuring points collect life-cycle data O ═ O from normal state to complete failure state1,O2,O3,O4}，Oi＝{Xi(1),Xi(2),Xi(3),...,Xi(n),...,Xi(984)}，Xi(n) is the measurement point tiData collected at the nth time. Fig. 3 shows the original vibration signal collected at four stations at n-1 (i.e., 10 minutes of operation).
And secondly, preprocessing the original data. The raw data O is preprocessed using equation (1), and as shown in fig. 4, the vibration signals of the four measuring points after being preprocessed when n is 1 (i.e. running for 10 minutes).
And thirdly, determining a fault prediction time node.
S3.1: and extracting fault early warning characteristics. Formula (2)) Middle fault early warning characteristic calculation function fFP(. r) is root mean square, and the root mean square value f at each time is calculatedFP(O (t)). Fig. 5 shows the fault early warning characteristic values corresponding to the four measuring points.
S3.2: and determining the fault prediction node, setting the early warning characteristic value FP in the early fault state to be 0.12, and calculating the fault prediction node TTP to be 665 according to the formula (3), namely corresponding to 6650 minutes.
And fourthly, constructing a fault growth trend.
S4.1-setting of the function f for calculating the characteristic of the trend of the failure in equation (4)FT(. r) is root mean square, and the root mean square value f at each time is calculatedFT(t)。
S4.2-to FFT(t) smoothing the sequence to establish a fault growth trend phi described by four measuring points1,Φ2,Φ3,Φ4And the failure growth trend described by the four measurement points after smoothing is shown in FIG. 6.
And S4.3, removing the fault growth trend of the non-trend and decreasing trend characteristics. M-K statistic Z for four failure growth trends calculated according to equation (5)1＝22.65，Z2＝-22.97，Z3＝-38.58，Z415.32. Thus, the failure growth trend set Φ for the significantly increasing trend characteristic with 95% confidencets＝{Φ1,Φ4And indicating that the increasing trend of the faults described by the 1 st measuring point and the 4 th measuring point have a significant increasing trend characteristic.
S4.4-calculate Phi according to equation (8)ts＝{Φ1,Φ4Trend in phi1And phi4Degree of correlation r140.5318 ≦ 0.8, indicating a failure growth tendency Φ1And phi4Fail-growth trend set Φ not exhibiting a high correlation and, therefore, free of redundant trend featuresrs＝Φts＝{Φ1,Φ4}。
Fifthly, fusing fault growth trend information
S5.1, calculating a fault growth trend set phi according to the formula (9)rs Medium 2 failure growth trend Φ1And phi4Weight ω of (d)1＝0.5965,ω4＝0.4035。
S5.2: using the formula (10) to phirsMean fault growth tendency Φ1And phi4Information fusion is carried out to obtain a fused fault growth trend sequence phiR＝ω1Φ1+ω4Φ4。
S5.3: for fused fault growth trend sequence phiRSmoothing is carried out to obtain the smooth fault growth trend phi*. Fig. 7 shows a curve of the increase trend of the failure after information fusion.
And sixthly, taking the fault growth trend set obtained in the fifth step as the input of a fault prediction model to predict the fault growth trend.
S6.1: with historical data phi*(1),...,Φ*(665) As input to the failure growth trend prediction model.
S6.2: method for predicting fault growth trend at T moment after early fault prediction point TTP (665) (6650 min) by using quadratic exponential smoothing prediction method
Figure GDA0003187991260000151
Fig. 8 shows a quadratic exponential smoothing prediction curve.
S6.3: setting a failure threshold FFT0.16, from the early failure prediction point TTP 665 (i.e. 6650 minutes) to the trend value after the 302 sample files have passed (i.e. 3020 minutes)
Figure GDA0003187991260000152
The point of failure TTF was calculated from (11) as 302 (time to failure as 3020 minutes).
The invention is not limited to the above alternative embodiments, and any other various forms of products can be obtained by anyone in the light of the present invention, but any changes in shape or structure thereof, which fall within the scope of the present invention as defined in the claims, fall within the scope of the present invention.

Claims (10)

1. A failure prediction method of multi-source trend information fusion is characterized in that: the method comprises the following steps:
s1, multi-source fault growth data acquisition: collecting raw data of a fault growth process at the same time interval by using a plurality of available measuring points;
s2, preprocessing the original data, and removing a trend item and an interference item;
s3, determining a fault prediction time node;
s4, constructing a fault growth trend: calculating a fault trend characteristic value of the original data at each moment, establishing a fault growth trend described by all measuring points, analyzing trend characteristics of the fault growth trend, and selecting a fault growth trend set with obvious monotone increasing; calculating the correlation degree of the fault growth trend in the set, and selecting a fault growth trend set without significant correlation;
s5, fusing fault growth trend information: fusing no significant related fault growth trend in the step S4 to obtain a redundancy-free fault growth trend set;
s6, using the redundancy-free fault growth trend set obtained in the step S5 as the input of the fault prediction model, and performing the following fault growth trend prediction:
a. predicting a fault trend characteristic value after the fault prediction time node;
b. and determining the time of the failure prediction time node to complete failure.
2. The multi-source trend information fusion fault prediction method of claim 1, characterized in that: in the step S1, the sensors deployed at each measurement point are used to collect raw data of a fault growth process, where the fault growth process is the whole fault growth process from a normal state to a complete failure state, and the specific implementation method is as follows:
preliminary deployment measurement point
Figure FDA0003187991250000011
NTCollecting the life-span data of the system from normal state to complete failure state for the total number of the measuring points
Figure FDA0003187991250000012
Wherein, OiFor the ith measuring point tiCollected fault evolution data.
3. The multi-source trend information fusion fault prediction method of claim 2, characterized in that: the step S2 is a specific implementation method for preprocessing the raw data and removing the trend term and the interference term as follows:
Figure FDA0003187991250000021
in the formula, Xi(n) is data OiData at time n,. mu.iAnd σiAre each XiMean and standard deviation of (n).
4. The multi-source trend information fusion fault prediction method of claim 3, wherein the fault prediction method comprises the following steps: the specific implementation method for determining the failure prediction time node in step S3 is as follows:
s31, calculating a fault early warning characteristic value of the original data at the time t, and calculating the fault early warning characteristic value of the data from the time 0-t by using a formula (2) to obtain a fault early warning characteristic value of the data O:
FP(t)＝fFP(O(t)) (2)
in the formula, FP (t) is a fault early warning characteristic value corresponding to the time t, O (t) is data collected at the time t, fFP() computing a function for the fault warning features;
s32, judging the fault early warning at the current moment, and calculating a fault prediction time node according to the formula (3):
Figure FDA0003187991250000022
in the formula (f)FP(O (t)) is a fault early warning characteristic value corresponding to the time t, FFPFor early warning threshold in early failure state, TTP is fFP(O (t)) and FFPAnd the relative error is less than or equal to 5 percent.
5. The multi-source trend information fusion fault prediction method of claim 4, wherein the fault prediction method comprises the following steps: the fault early warning characteristic calculation function is a time domain index, a frequency domain index or a time-frequency domain index;
the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, skewness, kurtosis, variance, a peak value, a standard deviation, a peak-to-peak value, average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index;
the frequency domain index is a power spectrum or a high-order statistic;
the time-frequency domain index is wavelet entropy or energy entropy.
6. The multi-source trend information fusion fault prediction method of claim 4, wherein the fault prediction method comprises the following steps: the specific implementation method of the step S4 is as follows:
s41, calculating the fault trend characteristic value of the original data at each moment by using the formula (4):
FT(t)＝fFT(O(t)) (4)
in the formula, FT (t) is a fault trend characteristic value corresponding to t time, O (t) is data collected at t time, fFT() computing a function for the fault trend signature;
s42, smoothing FT (t), and establishing the fault growth trend described by all measuring points
Figure FDA0003187991250000031
NTAs a total number of measured points, [ phi ]i＝{FT(1),FT(2),...,FT(NL) The ith measuring point is used for establishing a fault growth trend;
s43, analyzing trend characteristics of the fault growth trend, removing the measuring points corresponding to the decreasing fault growth trend with no trend characteristics, selecting the fault growth trend with obvious monotone increasing, and calculating the fault growth trend phi according to the formula (5)iM-K statistic Zi：
Figure FDA0003187991250000032
In the formula, ZiTo a failure growth tendency phiiM-K statistic of (N)LIs phiiNumber of samples of fault-evolution data collected in the process, SiTo a failure growth tendency phiiTest statistics, calculated according to equation (6):
Figure FDA0003187991250000033
wherein sign is a sign function when FT (n)1)-FT(n2) Sign (FT (n) when it is less than, equal to, or greater than zero1)-FT(n2) Are each-1, 0 or 1, NLIs OiThe total number of collected fault evolution data samples;
removing the trend-free features and the decreasing fault growth trend according to the formula (7), and keeping the fault growth trend set phi with the characteristic of obvious increasing trend with 95 percent of confidencets：
Figure FDA0003187991250000041
In the formula phiiThe increasing trend of the fault described for the ith station, ZiTo a failure growth tendency phiiM-K statistic of (N)TThe total number of the measuring points is;
s44, calculating phi according to the formula (8)tsMiddle tendency of phiiAnd phijDegree of correlation rijSelecting a set of fault growth trends without significant correlation:
Figure FDA0003187991250000042
in the formula phitsFor a set of failure growth trends with a significantly increasing trend characteristic, ΦiIs phitsThe ithThe tendency of the failure described by the measuring point to grow, i.e. phii∈Φts，NLIs OiThe total number of the collected fault growth data samples is removedtsMiddle | rijThe failure growth trend with the significant correlation of the value greater than 0.8 is obtained, and a failure growth trend set phi without redundancy trend characteristics is obtainedrs。
7. The multi-source trend information fusion fault prediction method of claim 6, wherein: the fault tendency characteristic calculation function is that the time domain index is a time domain index, a frequency domain index or a time-frequency domain index;
the time domain index is a mean value, a root mean square value, a square root amplitude value, an absolute mean value, skewness, kurtosis, variance, a peak value, a standard deviation, a peak-to-peak value, average power, a waveform index, a peak index, a pulse index, a margin index, a skewness index or a kurtosis index;
the frequency domain index is a power spectrum or a high-order statistic;
the time-frequency domain index is wavelet entropy or energy entropy.
8. The multi-source trend information fusion fault prediction method of claim 7, wherein: the specific implementation method of the step S5 is as follows:
s51, calculating a redundancy-free fault growth trend set phi according to the formula (9)rsWeight ω of each trend ini：
Figure FDA0003187991250000051
In the formula, ZiSet phi of growth trends for redundancy-free faultsrsMiddle ith tendency ΦiM-K statistic of (N)rsIs phirsMedian failure growth trend total;
s52, using equation (10) for phirsAnd carrying out information fusion on the medium fault growth trend to obtain a fused fault growth trend:
Figure FDA0003187991250000052
in the formula phirsFor a set of no-redundancy failure growth trends, phiRInformation fused set of fault growth trends, omegaiTo a failure growth tendency phiiWeight of (1), NrsIs phirsTotal number of median failure growth trends, Φi∈Φrs；
S53, merging the fault growth trend sequences phiRSmoothing is carried out to obtain the smooth fault growth trend phi*。
9. The method for predicting the failure of the multi-source trend information fusion of claim 8, wherein: the specific implementation method of the step S6 is as follows:
s61 in phi*(1),...,Φ*(TTP) data is historical data of failure growth trend prediction, failure growth trend is used as input to the prediction model, and TPP is an early failure prediction point;
s62, predicting the fault growth trend at the T moment after the TTP of the early fault prediction point by using a time series prediction method
Figure FDA0003187991250000053
S63, calculating the time to complete failure TTF from the early failure prediction point TTP using equation (11):
Figure FDA0003187991250000054
wherein TTP is an early failure prediction point, FFTFor a failure characteristic threshold at full failure,
Figure FDA0003187991250000061
the TTF predicts the TTP completion for the early failureTime of total failure, i.e. predicted value of time T elapsed from fault prediction point TTP
Figure FDA0003187991250000062
And a failure threshold value FFTThe relative ratio of (A) is less than or equal to 0.05.
10. The multi-source trend information fusion fault prediction method of claim 9, wherein: the time series prediction method is exponential smooth prediction, exponential function fitting or least square method.
CN201810866101.5A 2018-08-01 2018-08-01 Fault prediction method for multi-source trend information fusion Active CN109145773B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810866101.5A CN109145773B (en) 2018-08-01 2018-08-01 Fault prediction method for multi-source trend information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810866101.5A CN109145773B (en) 2018-08-01 2018-08-01 Fault prediction method for multi-source trend information fusion

Publications (2)

Publication Number Publication Date
CN109145773A CN109145773A (en) 2019-01-04
CN109145773B true CN109145773B (en) 2021-10-12

Family

ID=64798588

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810866101.5A Active CN109145773B (en) 2018-08-01 2018-08-01 Fault prediction method for multi-source trend information fusion

Country Status (1)

Country Link
CN (1) CN109145773B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI708197B (en) 2019-04-26 2020-10-21 國立成功大學 Predictive maintenance method for component of production tool and computer program product thereof
CN110222904B (en) * 2019-06-14 2023-01-31 谭晓栋 Monitoring point optimization method for fault quantitative propagation analysis
US20230229155A1 (en) * 2022-01-19 2023-07-20 Transportation Ip Holdings, Llc Inspection system and method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102567639A (en) * 2011-12-30 2012-07-11 南京航空航天大学 Method for evaluating reliability of aircraft engine aiming at competing failure
CN106874634A (en) * 2017-04-11 2017-06-20 中国人民解放军海军航空工程学院 Residual life Bayesian forecasting method based on inverse Gauss degradation model
CN107942943A (en) * 2017-12-25 2018-04-20 北京信息科技大学 A kind of high-end numerical control equipment state identification method based on Multi-source Information Fusion
EP3336636A1 (en) * 2016-12-19 2018-06-20 Palantir Technologies Inc. Machine fault modelling

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102567639A (en) * 2011-12-30 2012-07-11 南京航空航天大学 Method for evaluating reliability of aircraft engine aiming at competing failure
EP3336636A1 (en) * 2016-12-19 2018-06-20 Palantir Technologies Inc. Machine fault modelling
CN106874634A (en) * 2017-04-11 2017-06-20 中国人民解放军海军航空工程学院 Residual life Bayesian forecasting method based on inverse Gauss degradation model
CN107942943A (en) * 2017-12-25 2018-04-20 北京信息科技大学 A kind of high-end numerical control equipment state identification method based on Multi-source Information Fusion

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Prognostics and Health Monitoring of High Power LED;Sutharssan T 等;《Micromachines》;20121231;第78-11页 *
基于多源信息融合的雷达装备故障诊断决策研究;尚朝轩 等;《计算机测量与控制》;20110614;第19卷(第2期);第344-346页 *

Also Published As

Publication number Publication date
CN109145773A (en) 2019-01-04

Similar Documents

Publication Publication Date Title
Soualhi et al. Prognosis of bearing failures using hidden Markov models and the adaptive neuro-fuzzy inference system
CN109145773B (en) Fault prediction method for multi-source trend information fusion
Soualhi et al. Hidden Markov models for the prediction of impending faults
Shi et al. Rolling bearing initial fault detection using long short-term memory recurrent network
Bafroui et al. Application of wavelet energy and Shannon entropy for feature extraction in gearbox fault detection under varying speed conditions
CN109063308B (en) Health assessment method based on deep quantum learning
CN102252843B (en) Assessment method for rolling bearing performance variation
US10852357B2 (en) System and method for UPS battery monitoring and data analysis
CN111220387B (en) Vehicle bearing residual life prediction method based on multi-feature-quantity correlation vector machine
CN109376401B (en) Self-adaptive multi-source information optimization and fusion mechanical residual life prediction method
CA2875071A1 (en) Method and system for testing operational integrity of a drilling rig
CN109883691B (en) Gear residual life prediction method integrating kernel estimation and random filtering
Elforjani Diagnosis and prognosis of real world wind turbine gears
CN116226646B (en) Method, system, equipment and medium for predicting health state and residual life of bearing
Wen et al. Bearing remaining useful life prediction based on a nonlinear wiener process model
CN112683535B (en) Bearing life prediction method based on multi-stage wiener process
Martin-del-Campo et al. Dictionary learning approach to monitoring of wind turbine drivetrain bearings
CN111881594B (en) Non-stationary signal state monitoring method and system for nuclear power equipment
Shao et al. Gearbox vibration monitoring using extended Kalman filters and hypothesis tests
CN112861350A (en) Temperature overheating defect early warning method for stator winding of water-cooled steam turbine generator
Lv et al. A new feature extraction technique for early degeneration detection of rolling bearings
Guo et al. Research on remaining useful life of rolling bearings using EWT-DI-ALSTM
CN109165396A (en) A kind of equipment remaining life prediction technique of failure evolution trend
Liao et al. Machinery time to failure prediction-Case study and lesson learned for a spindle bearing application
CN114112390B (en) Nonlinear complex system early fault diagnosis 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