CN103995229B - A kind of feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method - Google Patents

Abstract

A kind of feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, first the vibration to motor, electric current and tach signal carry out data acquisition, then these signals are carried out feature calculation construction feature space, uses Feature Selection Algorithms to select the characteristic vector calculated for mahalanobis distance；Then calculate the mahalanobis distance of normal operating conditions motor, and construct the mahalanobis space of an expression motor normal operating conditions；The statistical parameter of the testing of electric motors signal reference normal operating conditions motor of unknown health condition is calculated mahalanobis distance, by the contrast with mahalanobis space, judges the health condition of testing of electric motors。The present invention, after being used for the motor signal under normal operating conditions training and be built into mahalanobis space, can effectively realize the motor of unknown duty is carried out health monitoring and abnormity diagnosis。

Description

A kind of feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method

Technical field

The invention belongs to For Diagnosing Faults of Electrical technical field, be specifically related to health monitoring and the abnormality diagnostic method of a kind of motor。

Background technology

Motor is widely used for the fields such as derived energy chemical, transportation, medical apparatus and instruments, office equipment。What the duty of motor was direct or indirect is related to these field devices reliabilities of operation。Therefore to its implement that health monitoring and abnormity diagnosis just seem very urgent。The fault of motor probably can be divided into mechanical fault, such as bearing fault, rotor fault and stator failure etc.；With the fault of electromagnetism class, such as the fault etc. of circuit, the fault of magnetic circuit system and insulation system。Current existing For Diagnosing Faults of Electrical technology contains amperometry, vibration analysis, Analysis of Acoustic Emission Signal and temperature analysis etc.。Have the disadvantage that when electrical fault pattern is not understood in detail, it is more difficult to realize the health monitoring of motor under unknown duty and abnormity diagnosis。

Summary of the invention

In order to overcome when electrical fault pattern not being understood in detail, the more difficult realization of existing fault diagnosis technology is to the health monitoring of motor under unknown duty and abnormality diagnostic deficiency, the present invention provides a kind of motor signal only needing to gather under normal operating conditions for after training and being built into mahalanobis space, can effectively realize the motor of unknown duty is carried out health monitoring and abnormality diagnostic feature based is chosen and the Method of Motor Fault Diagnosis of mahalanobis distance。

The technical solution adopted for the present invention to solve the technical problems is:

A kind of feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, said method comprising the steps of:

Step one, the motor signal gathered under normal operating conditions, including vibration, electric current and tach signal；

Step 2, described motor signal is carried out feature calculation obtain feature space, it is the calculating that vibration signal is carried out temporal signatures respectively, includes valid value, peak-peak, peak-to-peak value, kurtosis, meansigma methods, variance, standard variance, degree of skewness, crest factor and power；The simultaneously virtual value of calculating current and motor speed, and by one feature space of feature construction of described calculating；

Feature in step 3, the feature space that step 2 is obtained, uses the Feature Selection technology of minimal redundancy maximal correlation to select the characteristic vector calculated for mahalanobis distance；

Step 4, the characteristic vector that step 3 is obtained, carry out the statistical parameter of characteristic vector: the calculating of meansigma methods and standard variance, for the standardization of characteristic vector；To the characteristic vector after standardization, calculate its correlation matrix and inverse matrix thereof, and calculate mahalanobis distance, and construct the mahalanobis space represented under motor normal operating conditions；

Step 5, the characteristic vector that signals collecting through step one of the testing of electric motors of unknown health status and step 2 and step 3 are chosen calculate, the statistical parameter of the motor characteristic vector under utilization normal operating conditions, the characteristic vector of testing of electric motors is standardized, and using the inverse matrix of correlation matrix under normal operating conditions, the mahalanobis distance carrying out testing of electric motors calculates；

Step 6, the mahalanobis space mahalanobis distance of calculated for step 5 testing of electric motors and step 4 obtained compare, if the mahalanobis distance corresponding to testing of electric motors is in mahalanobis space, represent that testing of electric motors is working properly, otherwise then represent motor operation irregularity。

Further, in described step 3, feature f_iWith characteristic f_jBetween relevant information I (f_i,f_j) represent, feature f_iFrom relevant information I (c, the f between different motor c_i) represent, the standard of minimal redundancy refers to that the dependency between feature to reduce to minimum, usesRepresenting, wherein E represents the number of the sub-feature space E feature chosen in feature space S, and the standard of maximal correlation refers to that between feature and motor, dependency to keep maximum, and namely selected feature can represent the health condition of motor, usesRepresent, ratio forms may finally be used $\underset{E\⋐S}{\max }\{\underset{i\∈E}{\mathrm{\Σ}}I(c,f_i)/\frac{1}{\left|E\right|}\underset{i,j\∈E}{\mathrm{\Σ}}I(f_i,f_j)\}$ Or differential mode $\underset{E\⋐S}{\max }\{\underset{i\∈E}{\mathrm{\Σ}}I(c,f_i)-\frac{1}{\left|E\right|}\underset{i,j\∈E}{\mathrm{\Σ}}I(f_i,f_j)\},$ Each feature is marked, chooses the score feature construction more than zero and become characteristic vector。

Further, in described step 4, the characteristic vector composing training data P that the motor signal under normal operating conditions is chosen based on step 3 represents, p_ijRefer to and jth feature i & lt measurement is obtained data, j=1,2 ..., n, i=1,2 ..., m, n is the number of feature in characteristic vector, and m is the number of times that motor signal under normal operating conditions is gathered；

Calculate under normal operating conditions, the meansigma methods of jth featureAnd standard variance

$S_j=\sqrt{\frac{{{\mathrm{\Σ}}_{i=1}^m(p_{\mathrm{ij}}-\stackrel{\‾}{P_j})}^2}{m-1}};$

Then use the two statistical parameter to training data standardized calculation, as follows:

$z_{\mathrm{ij}}=\frac{p_{\mathrm{ij}}-\stackrel{\‾}{P_j}}{S_j}$

To the training data z after standardization, calculate correlation matrixWherein z_i=[z_i1,z_i2,...z_in]

Calculate the mahalanobis distance of training data afterwards:

${\mathrm{MD}}_i=\frac{1}{n}{z}_i{C}^{-1}{z}_i^T$

According to the mahalanobis distance value under the expression normal operating conditions calculated, set healthy threshold values, constitute mahalanobis space。

Beneficial effects of the present invention is mainly manifested in: after being used for the motor signal under normal operating conditions training and be built into mahalanobis space, can effectively realize the testing of electric motors of unknown duty is carried out health monitoring and abnormity diagnosis。

Accompanying drawing explanation

Fig. 1 is the flow chart of the motor health monitoring of the present invention and abnormality diagnostic method。

Fig. 2 is the mahalanobis distance schematic diagram under motor normal operating conditions。

Fig. 3 is the mahalanobis distance schematic diagram of testing of electric motors。

Detailed description of the invention

Below in conjunction with accompanying drawing, the invention will be further described。

With reference to Fig. 1, a kind of feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, comprises the following steps:

Step one, to the motor under normal operating conditions and testing of electric motors collection vibration, electric current and tach signal。

Step 2, the motor signal under the normal operating conditions collected is carried out property calculation obtain feature space, specifically: first, vibration signal is carried out the calculating of temporal signatures, includes valid value, peak-peak, peak-to-peak value, kurtosis, meansigma methods, variance, standard variance, degree of skewness, crest factor and power。Then, the virtual value of calculating current, it is built into feature space S together with motor speed；

Described feature space includes the virtual value of vibration signal, peak-peak, peak-to-peak value, kurtosis, meansigma methods, variance, standard variance, degree of skewness, crest factor and power, the virtual value of electric current and motor speed；

Feature in step 3, the feature space that step 2 is obtained, uses the Feature Selection technology of minimal redundancy maximal correlation to select the characteristic vector calculated for mahalanobis distance, has wherein applied to feature f in Feature Selection technology_iWith characteristic f_jBetween relevant information I (f_i,f_j) represent, and feature f_iFrom relevant information I (c, the f between different motor c_i) represent。The standard of minimal redundancy refers to that the dependency between feature and feature to reduce to minimum, availableRepresenting, wherein | E | represents the number of the sub-feature space E feature chosen in feature space S, and the standard of maximal correlation refers to that between feature and motor, dependency to keep maximum, and namely selected feature can represent the health condition of motor, usesRepresent, ratio forms may finally be used $\underset{E\⋐S}{\max }\{\underset{i\∈E}{\mathrm{\Σ}}I(c,f_i)/\frac{1}{\left|E\right|}\underset{i,j\∈E}{\mathrm{\Σ}}I(f_i,f_j)\}$ Or differential mode $\underset{E\⋐S}{\max }\{\underset{i\∈E}{\mathrm{\Σ}}I(c,f_i)-\frac{1}{\left|E\right|}\underset{i,j\∈E}{\mathrm{\Σ}}I(f_i,f_j)\},$ Each feature is marked, chooses the score feature construction more than zero and become characteristic vector。

The score of each feature of feature space S is calculated as shown in table 1 by the Feature Selection technology using minimal redundancy maximal correlation, choose to be divided into 5 features more than zero: power, crest factor, peak-peak, peak-to-peak value and motor speed, as being used for calculating the characteristic vector p of mahalanobis distance。

Feature	Mark
		Power P o	1.59
Crest factor CF	0.23
		Peak-peak Max	0.18
Peak-to-peak value PP	0.04
		Motor speed R	0.03
Other features	0

The score of each feature that table 1 calculates based on the Feature Selection technology of minimal redundancy maximal correlation

Step 4, the characteristic vector composing training data P that the motor signal under normal operating conditions is chosen based on step 3 represent, as shown in table 2。Table 2 show 10 initial datas of characteristic vector chosen。P_ijRefer to and jth feature i & lt measurement is obtained data, j=1,2 ..., n, n=5, i=1,2 ..., m, n is the number of feature in characteristic vector, and m is the number of times that motor signal under normal operating conditions is gathered。

Table 2 characteristic vector example

Calculate under normal operating conditions, the meansigma methods of jth featureAnd standard variance

$S_j=\sqrt{\frac{{{\mathrm{\Σ}}_{i=1}^m(p_{\mathrm{ij}}-\stackrel{\‾}{P_j})}^2}{m-1}}.$

Then use the two statistical parameter to training data standardized calculation, as follows:

$z_{\mathrm{ij}}=\frac{p_{\mathrm{ij}}-\stackrel{\‾}{P_j}}{S_j}$

To the training data z after standardization, calculate correlation matrixWherein z_i=[z_i1,z_i2,...z_in]

Calculate the mahalanobis distance of training data afterwards

${\mathrm{MD}}_i=\frac{1}{n}{z}_i{C}^{-1}{z}_i^T$

To the motor under normal operating conditions, the mahalanobis distance value calculated is all near 1, as shown in Figure 2。According to the mahalanobis distance value under the expressions normal operating conditions calculated, the threshold values of health being decided to be 2, just constitutes mahalanobis space, namely mahalanobis distance is from the scope between 0 to 2。

Step 5, to the testing of electric motors signals collecting through step one, the characteristic vector of step 2 and step 3 calculates, the statistical parameter of the motor characteristic vector under utilization normal operating conditions, it is standardized calculating to the characteristic vector of testing of electric motors, and use the inverse matrix of correlation matrix under normal operating conditions, the mahalanobis distance carrying out testing of electric motors calculates, as shown in Figure 3。

Step 6, the mahalanobis space mahalanobis distance of calculated for step 5 testing of electric motors and step 4 obtained compare, if the mahalanobis distance corresponding to testing of electric motors is in mahalanobis space, represent that testing of electric motors is working properly, otherwise then represent testing of electric motors operation irregularity, thus realize the health monitoring to motor and abnormity diagnosis。

Claims (3)

1. a feature based chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, it is characterised in that: said method comprising the steps of:

Step one, the motor signal gathered under normal operating conditions, including vibration, electric current and tach signal；

Step 2, described motor signal is carried out property calculation obtain feature space, it is the calculating that vibration signal is carried out temporal signatures respectively, includes valid value, peak-peak, peak-to-peak value, kurtosis, meansigma methods, variance, standard variance, degree of skewness, crest factor and power；The simultaneously virtual value of calculating current and motor speed, and by one feature space of feature construction of described calculating；

Feature in step 3, the feature space that step 2 is obtained, uses the Feature Selection technology of minimal redundancy maximal correlation to select the characteristic vector calculated for mahalanobis distance；

Step 4, the characteristic vector that step 3 is obtained, carry out the statistical parameter of characteristic vector: the calculating of meansigma methods and standard variance, for the standardization of characteristic vector；To the characteristic vector after standardization, calculate its correlation matrix and inverse matrix thereof, and calculate mahalanobis distance, and construct the mahalanobis space represented under motor normal operating conditions；

Step 5, the characteristic vector that signals collecting through step one of the testing of electric motors of unknown health status and step 2 and step 3 are chosen calculate, the statistical parameter of the motor characteristic vector under utilization normal operating conditions, the characteristic vector of testing of electric motors is standardized, and using the inverse matrix of correlation matrix under normal operating conditions, the mahalanobis distance carrying out testing of electric motors calculates；

Step 6, the mahalanobis space mahalanobis distance of calculated for step 5 testing of electric motors and step 4 obtained compare, if the mahalanobis distance corresponding to testing of electric motors is in mahalanobis space, represent that testing of electric motors is working properly, otherwise then represent motor operation irregularity。

2. a kind of feature based as claimed in claim 1 chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, it is characterised in that: in described step 3, feature f_iWith characteristic f_jBetween relevant information I (f_i,f_j) represent, feature f_iFrom relevant information I (c, the f between different motor c_i) represent, the standard of minimal redundancy refers to that the dependency between feature and feature to reduce to minimum, usesRepresenting, wherein | E | represents the number of the sub-feature space E feature chosen in feature space S, and the standard of maximal correlation refers to that between feature and motor, dependency to keep maximum, and namely selected feature can represent the health condition of motor, usesRepresent, ratio forms may finally be usedOr differential modeEach feature is marked, chooses the score feature construction more than zero and become characteristic vector。

3. a kind of feature based as claimed in claim 1 or 2 chooses the motor health monitoring with mahalanobis distance and abnormality diagnostic method, it is characterized in that: in described step 4, the characteristic vector composing training data P that motor signal under normal operating conditions is chosen based on step 3 represents, p_ijRefer to and jth feature i & lt measurement is obtained data, j=1,2 ..., n, i=1,2 ..., m, n is the number of feature in characteristic vector, and m is the number of times that motor signal under normal operating conditions is gathered；

Calculate under normal operating conditions, the meansigma methods of jth featureAnd standard variance

$S_j=\sqrt{\frac{{{\mathrm{\Σ}}_{i=1}^m(p_{\mathrm{ij}}-\stackrel{\‾}{P_j})}^2}{m-1}};$

Then use the two statistical parameter to training data standardized calculation, as follows:

$z_{\mathrm{ij}}=\frac{p_{\mathrm{ij}}-\stackrel{\‾}{P_j}}{S_j}$

To the training data z after standardization, calculate correlation matrixWherein z_i=[z_i1,z_i2,...z_in]

Calculate the mahalanobis distance of training data afterwards:

${\mathrm{MD}}_i=\frac{1}{n}{z}_i{C}^{-1}{z}_i^T$

According to the mahalanobis distance value under the expression normal operating conditions calculated, set healthy threshold values, constitute mahalanobis space。