KR20100112734A - On-site complex abnormal diagnosis method of induction motor - Google Patents

Abstract

PURPOSE: An on-site complex abnormal diagnosis method of an induction motor is provided to allow a user to efficiently maintain the induction motor by diagnosing various kinds of fault. CONSTITUTION: The characteristics of an induction motor are calculated from a vibration signal and a current signal(104). The feature values of a reference load rate are curved fitted and are calculated. The feature values of a breakdown characteristic map and induction motor are normalized(105). The feature of a group is extracted from the corrected feature values(106). The extracted feature is classified(107). The type of the problem is diagnosed in reference to the extracted value(108).

Description

On-site complex abnormal diagnosis method of induction motor

The present invention relates to a method for diagnosing a failure of an induction motor, and more particularly, in order to diagnose whether the induction motor is normally operated, calculating characteristic values detected from vibration signals and current signals of the induction motor, and calculating the calculated characteristic values. Through selective extraction process for the induction motor, it is possible to diagnose various kinds of failure states generated in the induction motor by diagnosing the fault condition of the induction motor from one or more characteristic values. It relates to a failure diagnosis method.

Induction motors may cause various induction motor defects such as dynamic eccentricity of the rotor, static eccentricity, rotor bar short circuit, stator winding interlayer short circuit, bearing failure, and so on. Various methods of diagnosing faults of induction motors that can be connected to accidents and accurately diagnose the state of induction motors have been studied.

The failure diagnosis method of the induction motor has been mainly performed for the early detection of equipment defects and the reduction of downtime by using the vibration technique, but the vibration technique alone makes it difficult to detect the motor defects early and accurately analyze the equipment. The necessity of precise diagnosis of rotating equipment has emerged by analyzing the characteristics of current, voltage and magnetic flux.

Insulation of stator winding of motor by acquiring data such as current, voltage, magnetic flux, temperature, vibration signal of induction motor used in rotating equipment (fan, pump, compressor, stirrer, etc.) or reciprocating equipment in industrial field New diagnostics that utilize various techniques to diagnose faults, breakage of rotors, instability of voids, and to monitor and correct bearing or shaft damage due to transients, to perform accurate diagnosis, to predict life, and to detect defects early. Technologies are being developed.

As such a diagnosis technique, a motor current signature analysis (MCSA) technique for diagnosing an induction motor using a frequency component of a current has been developed. The MCSA technique can remotely monitor the state of the motor without a sensor, and is effective for the rotor bar failure of the motor and useful for detecting other electrical failures. There is this. Therefore, an ESA (Electrical Signature Analysis) technique that uses a current and a voltage signal at the same time is additionally introduced to complement a diagnosis technique using only a current signal.

On the other hand, the electrical signals used for diagnosing the state of the induction motor include applied current and voltage, instantaneous applied power, motor void torque, and the like. These signals are used to predict the time axis, frequency axis, and time-frequency axis. In order to estimate the rotational speed, axial eccentric frequency detection and pole passing frequency detection are used. However, such an electrical signal analysis method requires a high resolution frequency analysis method, which requires a long time measurement and an excessive time for processing data.

In particular, when electrical signals such as stator current analysis are used, defects such as stator windings and rotor bar short-circuits that can occur in actual induction motors can be easily detected. There is a problem of being lowered.

In addition, when performing a fault diagnosis of an induction motor according to the prior art, it is necessary to determine whether or not a defect occurs for each defect, and there is a problem in that it is not possible to diagnose a complex defect state in which various defects are complex. In addition, it is difficult to accurately diagnose the abnormal operation state of an induction motor because it is not possible to distinguish the degree of defects simply by determining the presence or absence of a defect when diagnosing a failure.

Therefore, in the present invention, in the method of diagnosing the failure of the induction motor, the stator winding failure, through the analysis reflecting the measurement and the load factor for the vibration signal and the stator current signal to effectively detect the mechanical and electrical defects of the induction motor, The complex fault diagnosis method of the induction motor which can accurately detect the complex defects actually occurring in the induction motor such as rotor eccentricity and the bearing damage by one-time diagnosis, and can also diagnose the degree of the detected defects to provide.

In order to achieve the above object, the present invention provides the following configuration.

In the fault diagnosis method of the induction motor, the method comprises the steps of: a) creating a fault characteristic map comprising a characteristic value for a steady state and a fault condition according to a preset reference load ratio; b) calculating characteristic values of the diagnostic induction motor from the vibration signal and the current signal of the diagnostic induction motor; c) curve fitting feature values for the reference load factor to calculate corrected feature values corresponding to the actual load factor of the diagnostic induction motor; d) normalizing the feature values of the fault feature map and the feature values of the diagnostic induction motor; e) extracting feature values forming a cluster from among the corrected feature values reflecting the actual load rate so as to distinguish the failure types; f) classifying a cluster of the extracted feature values according to whether the cluster of extracted feature values is included within a predetermined threshold value from the calculated feature value of the diagnostic induction motor; g) detecting a cluster of the maximum ratio among the clusters of the classified feature values and a cluster whose ratio and the difference between the maximum ratio cluster are within a predetermined diagnostic parameter and diagnosing a failure type corresponding to the detected cluster; An on-site complex failure diagnosis method of an induction motor is provided.

In this case, the reference load ratio of the step a) provides an on-site combined failure diagnosis method of the induction motor, characterized in that the load ratio of 0%, 25%, 50%, 75%, 100%.

In addition, the kind of failure state of step a) is one or more defects selected from the group consisting of a dynamic eccentricity of the rotor, a static eccentricity of the rotor, a short circuit of the rotor bar, a short circuit between the stator windings, and a bad bearing. An on-site complex failure diagnosis method of an induction motor is provided.

In addition, the fault conditions of step a) are classified into light and heavy conditions according to the degree of the fault condition, and a fault feature map in which feature values for each state are independently stored is generated. -Site complex fault diagnosis method is provided.

As described above, the on-site complex failure diagnosis method of the induction motor according to the present invention has the following effects.

First, since the vibration signal and the stator current signal can be detected and the analysis can be performed by reflecting the actual load rate of the induction motor corresponding to the diagnosis target, accurate diagnosis can be made for both mechanical and electrical defects of the induction motor. .

Second, there is an effect to enable a complex diagnosis that accurately diagnoses the failure of each of a plurality of defects by the one-time diagnosis method for the failure state of the induction motor in which various defects are generated in combination.

Third, it is possible to determine whether the induction motor defects occur as well as the degree of defects such as light and medium defects, so that the maintenance and repair of the induction motor can be effectively performed.

Fourth, by providing a fault diagnosis method through a simple feature extraction and analysis process, the fault diagnosis of the induction motor can be accurately performed even with a simple facility, thereby improving the operational reliability of the induction motor and minimizing the ripple effect due to the failure. There is.

In the present invention for achieving the above object, in the complex failure diagnosis method of the induction motor, in order to detect the mechanical and electrical defects of the induction motor, the characteristic value calculated from the vibration signal and the stator current signal, and the load factor is reflected analysis By extracting a cluster of feature values corresponding to a predetermined diagnostic parameter from a cluster of feature values corresponding to a predetermined threshold value through the detection of complex defects generated in the induction motor, as well as through a predetermined data provided in advance The present invention provides a complex failure diagnosis method of an induction motor capable of determining a degree of a defect by comparing the detected defect data.

In the on-site complex fault diagnosis method of an induction motor according to the present invention, the vibration signal and the stator current signal are simultaneously measured from the induction motor for fault diagnosis, and the characteristic values for the vibration signal and the current signal for fault diagnosis are calculated. The step is performed first. Then, the calculated feature values reflect the current operating state of the induction motor, and after extracting predetermined feature values determined to be suitable for fault diagnosis, the feature values of the pre-calculated fault feature map and the extracted feature values By comparing this, it is determined whether a fault occurs in the induction motor based on the threshold and the setting range of the diagnostic parameter.

Therefore, it is possible to simultaneously compare and analyze the calculated feature values and the data of feature values previously prepared for each defect according to the threshold and the setting range of the diagnostic parameter, and to turn on the induction motor that can diagnose a complex defect in one diagnosis. -Site complex fault diagnosis method is provided.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 of the accompanying drawings is a flowchart sequentially illustrating a process of actually performing the on-site complex failure diagnosis method of an induction motor according to the present invention. As shown in FIG. 1, the on-site complex failure diagnosis method of an induction motor according to the present invention includes a data acquisition step, a step of calculating a feature value using vibration and current signals, a normalization step, a step of extracting a feature value, It is configured to go through the classification step of the feature value, the failure determination step.

Prior to the detailed description of each step in the on-site complex failure diagnosis method of the induction motor according to the present invention, the feature values calculated for defect determination are as follows.

The feature value refers to a parameter of a time domain or a frequency domain of an induction motor. The feature values for determining a defect in the present invention are characterized by a slip value and a rotation speed-dependent feature value (hereinafter referred to as 'first group'). .), It is classified into two groups, which are divided into characteristic values independent of slip and rotation speed (hereinafter referred to as 'second group').

The feature value corresponding to the first group is divided into a feature value calculated from the detected vibration signal and a feature value calculated from the current signal, and the fault frequency of the bearing defect is calculated from the vibration signal, and from the current signal, respectively. Calculate the fault frequency of stator winding interlayer short circuit, rotor bar short circuit, and eccentricity fault.

Since the vibration signal has the most accurate diagnosis rate for actual bearing defects, the defect frequencies of the inner and outer rings of the bearing are calculated to diagnose the defects of the inner and outer rings of the bearing.

The fault frequency of the bearing inner ring

The fault frequency of the bearing outer ring is

(n: number of bearings, f _r : rotational frequency (Hz))

On-site complex failure diagnosis method of the induction motor according to the present invention is mainly for the failure diagnosis of low-pressure induction motor, the bearing fault frequency calculation formula of the inner and outer ring is a calculation formula used when the number of ball bearings 6 ~ 12, Considering the general view of induction motors, they are commonly available.

On the other hand, as a current signal, short circuit defects between stator windings, rotor bar short circuits, and eccentric diagnosis are performed.

The fault frequency of the stator winding interlayer short circuit fault is

(f _o : supply frequency, l: 1, 2, 3, ……, k: 1, 3, 5, ……, p: pole pairs, s: slip)

In this case, the defect frequency of the stator winding interlayer short-circuit defect is determined by l and k. However, considering the data processing speed and memory capacity, predetermined values (experimental results, k values up to 1, 3, 5, l It is desirable to set the value up to 1, 2, 3, 4, 5 until the calculation is sufficient for fault diagnosis.

In addition, the defect frequency of the rotor bar short circuit defect is

(f _o : supply frequency, s: slip, k: 1, 2, 3, …… (typically k = 1))

As in the above-described short circuit of the stator winding interlayer, it is preferable to specify a value of k. In the preferred embodiment of the present invention, k = 1 is set because k is a signal of a defect frequency.

The defect frequency of the eccentric defect is

(f _o : supply frequency, s: slip, n _ws : 1, 2, 3, ……, R: number of rotor bar slots, p: pole pairs, n _d0 : 0 (static eccentricity), 1 (dynamic eccentricity))

According to the test results performed according to the present invention, it can be seen that the defects can be sufficiently accurate even if the nws value is calculated only up to 1, 3, and 5.

Meanwhile, in the on-site complex fault diagnosis method of the induction motor according to the present invention, the feature values in the time domain among the feature values corresponding to the second group are mean, standard deviation, skewness. ), Kurtosis, RMS, shape factor, crest factor, and the characteristic values of the frequency domain are frequency center, mean square frequency, It corresponds to a root mean square frequency, a variance frequency, and a root variance frequency. Briefly look at the calculation formula for each feature value of the second group according to the present invention.

The average is a value representing the entire time series signal and is expressed as follows.

The mean is based on the information contained in all observations, but corresponds to a feature that is sensitive to the effects of extreme observations, such as an impulse waveform.

Although variance and standard deviation can be used as a measure of the dispersion of time series signals, the standard deviation is used in the present invention because the standard deviation corresponds to a meaningful characteristic value compared to the variance, and the standard deviation is calculated by the following equation. .

Skewness represents the degree of symmetry and the direction of the probability density function distribution of observed values for a time series signal. Distortion is calculated by the Pearson method (Equation 1), or the method using the ratio of the standard deviation to the mean (Equation 2), obtained by dividing the difference between the mean and the median (or mode) by the standard deviation. The formula for each is as follows.

: 제1식, (mode는 최빈값, median은 중앙값을 의미한다.)

: Formula 1 (mode means mode, median means median.)

: 제2식

: 2nd meal

Kurtosis is a measure of the sharpness of the probability density function distribution of a time series signal and represents the quantitative structure of the data.

RMS is used as a measure for quantifying vibration amplitude, which is a characteristic that indicates the magnitude of vibration, since the change in time is taken into account and displays an amplitude directly related to the amount of energy representing the breaking force of vibration. The effective value is obtained from the following formula.

Shape factor represents the ratio of the vibration mean to the effective value of vibration,

It is represented as

Crest factor represents the ratio of peak to effective value of vibration and is used to identify mechanical defects such as bearing or gear defects. The definition of crest factor is as follows.

In the meantime, the frequency center among the characteristic values of the frequency domain means the center of the spectral density when the discrete time series signal is viewed in all frequency bands. The relation of the frequency center with respect to the continuous signal is as follows.

: 연속 신호에 대한 주파수 중심의 관계식 (s(f)는 파워 스펙트럼을 의미한다.)

: Frequency center relation for continuous signal (s (f) means power spectrum)

The frequency center relation for the discrete signal is as follows.

: 이산 신호에 대한 주파수 중심의 관계식

: Frequency-centered relationship to discrete signals

The root mean square frequency is calculated for the continuous and discrete signals, respectively,

: 연속 신호에 대한 관계식

= Relational expression for continuous signal

: 이산 신호에 대한 관계식

= Relational expression for discrete signals

The effective frequency is a feature value corresponding to the square root of the square mean frequency,

Dispersion frequency is a value representing the variance between frequencies of different components around a center frequency in the entire spectrum, and is calculated by the following equation.

또는

or

The standard deviation frequency corresponds to the square root of the variance frequency,

에 의하여 계산된다.

Calculated by

As described above, the feature values corresponding to the first group and the feature values corresponding to the second group are calculated from current signals and vibration signals obtained from the induction motor.

However, although the characteristic values corresponding to the second group among the characteristic values calculated in the on-site complex failure diagnosis method of the induction motor according to the present invention do not change depending on the rotational speed and the slip of the induction motor, the first group The characteristic values belong to are dependent on the rotational speed and slip of the induction motor and change as these values change.

Therefore, in order to calculate the feature values of the first group, calculation values for the rotational speed and the slip of the induction motor are required. In the present invention, the rotational speed calculation using the rotor slot harmonics (RSH) of the current signal is required. Use the method.

The formula for calculating the rotation speed is

(f_r: 회전 속도(Hz), p: 극 쌍의 수, Z: 회전자 슬롯의 수, α=±k (k: 양의 정수), f_o: 공급 주파수, f_sh: 슬롯 조화 주파수)이다.

(f _r : rotational speed (Hz), p: number of pole pairs, Z: number of rotor slots, α = ± k (k: positive integer), f _o : supply frequency, f _sh : slot harmonic frequency) to be.

If no load, fr = fo, so using the above formula,

(f_sh0: 무부하일 때 슬롯 조화 주파수)이다.

(f _sh0 : slot harmonic frequency at no load).

When the slip frequency fs is calculated through the slot harmonic frequency, the slip frequency is calculated by the following equation.

In this case, the α value uses α = −3, which is a value at which a slot harmonic frequency exists at all loads.

In FIG. 2, the slot harmonic frequencies corresponding to the load ratios 0, 25, 50, 75, and 100% at α = -3 in the steady state are shown. (Induction motor is 7.5㎾ 4 pole: Z = 28, p = 2, f _o = 60Hz)

Based on the characteristic values corresponding to the first group and the second group as described above, a preferred embodiment of the on-site complex failure diagnosis method of the induction motor according to the present invention will be described in detail with reference to the flowchart of FIG. do.

As shown in FIG. 1, the method for diagnosing on-site complex failure of an induction motor according to the present invention is configured such that first, a step 101 of acquiring a current signal and a vibration signal from an induction motor is performed. In this case, all the data on the three-phase current of the general induction motor can be obtained and analyzed.However, in the case of feature extraction for motor fault diagnosis, only one-phase current signal can be used for sufficient analysis. It can be configured to obtain only the data for the one-phase current signal under consideration.

Next, the rotation speed and slip of the induction motor to be diagnosed are calculated by using the one-phase current signal obtained through step 101 (102), and the experiment is performed in advance on the state of the induction motor using the calculated slip and rotation speed values. In this case, the feature values reflecting the load factor are calculated and extracted from the parameters of the failed feature map. In this case, the feature values are divided into a first group and a second group according to whether they are independent of slip and rotation speed, The feature values of the second group may be configured to include feature values independent of the slip and rotation speed.

The fault characteristic map is data prepared in advance for parameters corresponding to a predetermined characteristic value in order to diagnose a fault of the induction motor, and conducts experiments reflecting a load ratio according to a fault condition of the induction motor, thereby performing respective vibration signals and current signals. It means to obtain the feature value for and store it as data.

In a preferred embodiment of the present invention, the induction motor and six defect rotors in a steady state (Bowed Rotor, Rotor Misalignment), the rotor's static unbalance (Mass Unbalance), rotor bar short circuit (Broken Rotor Bar), Fault characteristics maps for 13 states were prepared by dividing the stator winding between short-circuit (Stator Winding Fault) and broken bearing (Broken Bearing) into two states (light and medium), respectively. The load rate was divided into 0%, 25%, 50%, 75%, and 100%.

As described above, specific examples of the light and medium conditions of the faults, that is, the degree of defects of the six defects are shown in Table 1 below.

Electric motor type

Defect History
% Defective
Normal electric motor

Bowed Rotor

0.4mm eccentric
80.0
Bowed Rotor

0.3mm eccentric
64.0
Mass Unbalance of Rotor

50g × 2ea
Mass Unbalance of Rotor

50g × 1ea
Rotor Misalignment

0.4mm eccentric
80.0
Rotor Misalignment

0.3mm eccentric
56.0
Broken Bearing

3mm hole (1ea)
Broken Bearing

6mm hole (1ea)
Broken Rotor Bar

4ea
14.3
Broken Rotor Bar

2ea
7.1
Stator Winding Fault

5 turn (total 63turns)
7.9
Stator Winding Fault

10turn (total 63turn)
15.9

As shown in Table 1, the state for each defect is set to light and medium state, the defect details in Table 1 are specific details of the defect in the set state, the degree of defect is set to 0% normal state At this time, the degree of the defect is expressed as a percentage, and it is left blank that the degree of the defect cannot be indicated.

For the accurate fault diagnosis related to the fault feature map created according to the load rate, a fault feature map that stores the feature values accurately reflecting the load rate of the induction motor to be diagnosed is required.However, experiments are conducted for all states according to each load rate. In the present invention, inefficiency of obtaining and storing data is generated, and thus, data for a predetermined reference load rate (data for a load rate of 0%, 25%, 50%, 75%, and 100% in a preferred embodiment of the present invention) may be obtained. It provides a fault feature map comprising a), and is configured to correct a change in the magnitude of the feature value according to the actual load ratio from the fault feature map. In order to perform the correction according to the load ratio, the present invention provides a correction method through curve fitting. As shown in FIG. 3, as an example of such magnitude correction, an unknown load ratio is obtained by curve fitting a feature value for an effective value of a current signal. It is shown that the size correction for can be performed.

FIG. 4 is a block diagram illustrating the characteristic value calculation step 103 reflecting the load ratio as described above in more detail. In the on-site complex failure diagnosis method of an induction motor according to the present invention, a plurality of times are used for accuracy of diagnosis. In order to calculate the characteristic value reflecting the load factor from the characteristic values of the fault characteristic map measured and stored over, from the characteristic values for the reference load rate stored in the fault characteristic map, the size correction process as shown in FIG. It is configured to calculate a corresponding feature value.

Referring to the size correction process according to the load ratio in more detail, as shown in Figure 4, in the on-site complex failure diagnosis method of the induction motor according to the present invention, the characteristic value of the reference load factor for the 20 times the failure feature map In the case of storing to, the curve fitting and corrected feature values are not calculated for all experiments. After performing only one curve fitting, the feature values reflecting the actual load rate for all the experiments are extracted. do. That is, after extracting (401) feature values of the load factor closest to the actual load rate from the fault feature map, an average value (A) of the extracted feature values is calculated (402), and the mean value of the feature values stored in the fault feature map. The curve-fitted result showing the average tendency according to the change of the reference load ratio which they have is calculated (411), and the average value (B) of the feature values corresponding to the actual load ratio is calculated (412). Next, the ratio B / A of the average value A to the feature value of the load factor closest to the calculated real load rate and the average value B of the feature value corresponding to the actual load rate is calculated 421, and in step 401 By multiplying the ratio (B / A) for each characteristic value corresponding to the reference load ratio closest to the actual load ratio of (403), the feature values whose magnitude correction is performed on the load ratio, that is, the characteristic values reflecting the load ratio Extract them (404)

On the other hand, after calculating the characteristic value reflecting the load factor from the fault characteristic map 103 is performed, the characteristic values of the induction motor for diagnosis are calculated through the obtained data as in the case of the fault characteristic map prepared in advance. Step 104 is performed to normalize the feature values reflecting the calculated feature values of the diagnostic induction motor and the load factor extracted from the fault feature map. This normalization step is to solve the size difference of the characteristic values generated according to the induction motor type so that the diagnosis of various induction motors can be performed according to the fault feature map prepared in advance.

The normalized feature values are subjected to a feature value extraction step 106 of extracting features suitable for fault diagnosis of an induction motor by kernel principal component analysis (KPCA), and a defect through feature value extraction by the kernel principal component analysis. Only feature values that are easy to diagnose are extracted, and the feature values selected for fault diagnosis are classified through the k-NN classifier. (107)

In relation to this feature value extraction step, Kernel Principal Component Analysis (KPCA) used in the present patent for feature value extraction is a Principal Component Analysis (PCA) extension technique, which is a principal component analysis (Principal Component analysis). Analysis: PCA) is a useful statistical technique applied in various fields such as facial recognition and image compression. In this principal component analysis, the principal component (PC) is ordered such that the k th principal component has the largest variance at k th among all principal components, and the k th principal component is orthogonal to the k-1 th principal component. It can be described in the direction of maximizing the deviation of the projection.

A common approach is to use the first few principal components in the data analysis that include most of the changes in the original data set, and assume the remaining principal components as residual noise in the data. For a given p-dimensional dataset X, m principal axes T ₁ , T ₂ ,... , Ｔ _m (1 ≦ m ≦ p) is an orthogonal axis with the largest dispersion in the projected space. In general, T ₁ , T ₂ ,.. , Ｔ _m Is given from the m eigenvectors of the sample covariance matrix S. The covariance matrix is given by

, μ는 평균, Ｎ 은 샘플의 수이다. 고유치 문제는here

is the mean and N is the number of samples. The eigenvalue problem

Where λ _i is the i-th largest eigenvalue of S.

Thus, the m principal components of a given measurement data vector are given by The principal components are in the form of the product of the eigenvectors obtained from the eigenvalues and the original data X in the covariance matrix S and the y vector is obtained.

The m principal components of x become independent of each other in the projected space.

사이에 비선형 관계를 분리시키기 위한 목적으로 사용되며, 상호상관은 비선형 공간의 입력 대신에 선형 특징 공간의

로 표현될 수 있다. 커널 주성분 분석의 아이디어는 입력 벡터

를 고차원

특징 공간으로 사상시키고 그리고나서

에서 PCA를 계산한다.

안에 선형 PCA는

안에 비선형 주성분 분석과 관계가 있다.

를

에 사상시킴을 통해 차원은 훈련 샘플의 수보다 차원이 높아지게 된다. Meanwhile, Kernel Principal Component Analysis (KPCA) is a method of generalizing linear principal component analysis using a kernel function which is a nonlinear approach. Data Set Given Principal Component Analysis Via Diagonalized Correlation Matrix

Used for the purpose of separating nonlinear relationships between

It can be expressed as. Idea of Kernel Principal Component Analysis Input Vector

High dimension

And then map it into the feature space

Calculate the PCA.

Linear PCA inside

It is related to nonlinear principal component analysis.

To

By mapping to the dimension, the dimension is higher than the number of training samples.

KPCA can be solved by eigenvalues

는

의 상호 상관 행열이다.

는

의 0 이 아닌 고유치중의 하나이다.

는 고유벡터이다. 고유치들은

와

의 조건을 만족해야 한다. 상기의 식으로부터 얻어진 가장 큰 고유치

에 따른 고유벡터

은 특징공간

상에서 주성분(PC : Principal Component)이 된다. 그리고 가장 작은 고유치

에 따른 고유벡터

는 마지막 주요소가 된다.At this time,

Is

Is the cross-correlation matrix of.

Is

One of the nonzero eigenvalues of.

Is the eigenvector. Eigenvalues

Wow

The conditions of Largest eigenvalue obtained from the above equation

Eigenvectors According to

Silver feature space

It becomes the principal component (PC: Principal Component) in the phase. And the smallest eigenvalue

Eigenvectors According to

Becomes the last major element.

Feature extraction by kernel principal component analysis is as follows.

으로부터 커널 함수를 이용하여 계산한다.(단계 S₁)First, the original data (feature value)

Calculate using the kernel function from. (Step S ₁ )

상의 중심으로 옮긴다. (단계 S₂)Next, the coordinate space of the data calculated by the kernel function is increased

Move to the center of the statue. (Step S ₂ )

Perform normalization of the data (step S ₃ )

의 고유치를 계산하고,(단계 S₄), 고유치를 크기순으로 나열한 다음(단계 S₅), 고유치에 기초한 주성분을 선택하는(단계 S₆) 과정을 통하여 커널 주성분 분석에 의한 특징 추출을 수행한다.procession

The feature extraction by kernel principal component analysis is performed by calculating the eigenvalue of (step S ₄ ), arranging the eigenvalues in order of magnitude (step S ₅ ), and then selecting the principal component based on the eigenvalue (step S ₆ ). .

5 and 6 of the accompanying drawings are distribution charts showing the distribution of feature values before and after performing the kernel principal component analysis. As shown in FIG. 5, the characteristic values are not clearly clustered before the kernel principal component analysis is performed. However, as shown in FIG. 6, the feature values extracted after the kernel principal component analysis may confirm the steady state and the distribution of feature values that are clearly clustered for each defect.

On the other hand, the feature values extracted through the feature extraction process as described above are based on the data of the fault feature map prepared in advance for the steady state and each defect state (including the mild and medium states of the defect). A feature value classification step 107 is carried out so that the feature values calculated from the current and vibration signals can actually be classified into clusters of corresponding defects. In order to perform the classification step of the feature value, the present invention uses a k-NN classifier, and in the hyper-sphere having a predetermined range defined in advance the clusters for the steady state and each defect state belonging to After sorting and screening, a defect determination step 108 is performed to determine one or more defects (complex defects) of the induction motor. In the above-described step, the tool has a radius corresponding to a predetermined threshold, and provides information on various kinds of defects corresponding to the current state in space. Accordingly, in the present invention, in order to effectively detect one or more defects that are actually occurring in the induction motor, that is, a complex defect, a predetermined threshold value k for the bulb is set, and the characteristic values of the respective clusters are determined. As the distance difference between the feature values of the induction motor under test selects all of the feature values corresponding to the predetermined threshold value k for the bulb, the selected feature values can be classified into groups.

For example, the five classification items classified as a, b, c, d, and e may be extracted from the data (failure feature map) prepared in advance from the previously generated data (fault feature map) to the closest 100th value. The threshold value may be set as a distance or a number of feature values to be extracted.) By selectively extracting feature values close to a distance with respect to an input based on the set threshold value k = 100, the threshold values may be classified according to each cluster.

On the other hand, in the on-site complex failure diagnosis method of the induction motor according to the present invention, by setting a predetermined diagnostic parameter (P) to detect all kinds of failures that are determined to be a fault among the extracted and classified feature values Configure the system to perform accurate complex fault diagnosis.

The diagnostic parameter P is a parameter set to detect complex defects from the classified feature values and is a predetermined range corresponding to the diagnostic parameter from the ratio of the maximum detected cluster among the clusters of the classified feature values on the k-NN classifier. And selecting the clusters included in the overlapping to determine the defect. This diagnostic parameter (P) is for detecting a cluster of feature values determined to be actual failures among various feature values, and is set to a value selected from a range corresponding to 5% to 20%, preferably 5% to 10%. do.

For example, if the diagnostic parameter (P) is defined as 10%, if the proportion of the data classified through the above embodiment is classified as 50% a, 42% b, 8% c, k-NN The result of the classifier is judged as a and b. That is, since the diagnostic parameter (P) 10% range is applied to 50% of a, the maximum cluster, a, b is a deficiency of the induction motor, including b corresponding to 40% or more. Judging by

On the other hand, in the above example, if a is 35%, b is 30%, c is 28%, d is 7%, and the diagnostic parameter (P) is 10%, a, b, and c are determined to be induction motor defects. do.

Therefore, as described above, by setting the threshold value and the diagnostic parameter for the initial tool, the clusters included in the predetermined threshold value are classified, and the diagnosis parameter is applied to the ratio of the classified clusters, thereby making it possible to diagnose a complex failure of the induction motor. . That is, when a cluster corresponding to a bearing failure and a defect of a rotor bar short circuit is located within a predetermined threshold and satisfies a range of predetermined diagnostic parameters, failure diagnosis of the two defects can be performed simultaneously, so that a single diagnosis can be made. As a result, the diagnosis of complex defects can be performed, thereby improving the reliability of the diagnosis.

While the invention has been described with reference to the preferred embodiments, those skilled in the art will appreciate that modifications and variations of the elements of the invention may be made without departing from the scope of the invention. In addition, many modifications may be made to particular circumstances or materials without departing from the essential scope of the invention. Therefore, the invention is not limited to the details of the preferred embodiments of the invention, but will include all embodiments within the scope of the appended claims.

1 is a flowchart sequentially illustrating a process in which an on-site complex failure diagnosis method of an induction motor according to the present invention is performed.

2 is a graph of slot harmonic frequency according to load ratio.

3 is a curve-fitted graph to reflect changes in feature values with load ratios.

Figure 4 is a block diagram illustrating a process of extracting a feature value reflecting the load factor in the present invention.

5 is a distribution chart of feature values before kernel principal component analysis (KPCA).

6 is a distribution diagram of feature values after kernel principal component analysis (KPCA).

Claims (6)

a) creating a fault characteristic map consisting of feature values for steady state and fault conditions according to a preset reference load rate; b) calculating characteristic values of the diagnostic induction motor from the vibration signal and the current signal of the diagnostic induction motor; c) curve fitting feature values for the reference load factor to calculate corrected feature values corresponding to the actual load factor of the diagnostic induction motor; d) normalizing the feature values of the fault feature map and the feature values of the diagnostic induction motor; e) extracting feature values forming a cluster so as to distinguish between types of failures among the corrected feature values reflecting the actual load ratio; f) classifying a cluster of the extracted feature values according to whether or not the cluster of extracted feature values is included within a predetermined threshold from a calculated feature value of the diagnostic induction motor; g) detecting a cluster having a maximum ratio among the clusters of the classified feature values and a cluster whose ratio and the difference between the maximum ratio cluster are within a predetermined diagnostic parameter P and diagnosing a failure type corresponding to the detected cluster; On-site complex failure diagnostic method of the induction motor, characterized in that made. The on-site complex failure diagnosis method of claim 1, wherein the reference load rate in step a) is 0%, 25%, 50%, 75%, 100%. The method of claim 2, wherein the step c) extracts feature values of the reference load rate that is closest to the actual load rate from the failure feature map, calculates an average value A thereof, and curve fitting the average value of the feature values for each reference load rate. By calculating the average value (B) of the feature values according to the actual load rate, and calculating the corrected feature values corresponding to the actual load rate of the diagnostic induction motor from the ratio (B / A) How to diagnose on-site complex failure of a motor. The method of any one of claims 1 to 3, wherein the type of failure state in step a) includes a dynamic eccentricity of the rotor, a static eccentricity of the rotor, a shorter rotor bar, a short circuit between stator windings, and a poor bearing. On-site complex failure diagnosis method for an induction motor, characterized in that at least one fault selected from the fault. The induction method according to claim 4, wherein the failure states of step a) are divided into light and heavy states according to the degree of the failure state, and a failure feature map in which feature values for each state are independently stored is generated. How to diagnose on-site complex failure of a motor. The method of any one of claims 1 to 3, wherein the diagnostic parameter (P) corresponds to a preset value in the range of 5% to 10%.

Images