JP2014194727A - Rotary machine quality diagnostic system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a system capable of diagnosing the presence or absence of abnormality occurring in an operating rotary machine without stopping the rotary machine while a load is connected.SOLUTION: A rotary machine quality diagnostic system diagnoses quality of a rotary machine during operation by: measuring current flowing to each phase of a normal rotary machine that operates by being connected to a load, with respect to each of differently-sized loads; obtaining a feature quantity of each load; obtaining a basic line of a region where the feature quantities are distributed; defining a probability density function of an index to be used for determining the quality of the rotary machine, as a function of distance from the line; deciding a value of a standard deviation in the probability density function so that the index takes a specific value when the distance between the feature quantity and the line becomes a certain value; storing this value in a computer; obtaining the distance between a feature quantity obtained in diagnosing the rotary machine and the line when the feature quantity is obtained; using a value of the distance and the preliminarily-obtained value of the standard deviation to calculate a failure probability showing a determination index; and setting a threshold to the value.

Description

  The present invention relates to a technique for diagnosing the presence or absence of an abnormality that occurs in a rotating machine such as an electric motor or a generator operating with a load connected.

  Conventionally, for example, in the case of diagnosing a winding abnormality of a rotating machine, a technique using an impulse test has been reported (see Non-Patent Document 1).

  Electronic Control International Co., Ltd. Impulse Winding Tester DXW-01, 05 Instruction Manual

  However, the conventional method (Non-Patent Document 1) is a technique for diagnosing the rotating machine in a stopped state, and has a problem that it cannot be applied to an actually operating rotating machine.

The present invention aims to provide a system capable of diagnosing the presence or absence of an abnormality occurring in a rotating machine operating with a load connected without stopping the rotating machine, and achieves at least a part of this object. The following measures were taken for this purpose.
The rotating machine quality diagnosis system of the present invention is:
The current flowing in each phase of a normal rotating machine operating with a load connected is measured for each load of various magnitudes, and feature quantities are obtained for various load magnitudes. The area in which the quantity is distributed is obtained, the reference line (feature quantity distribution line) of the area in which the feature quantity is distributed in the normal rotating machine is obtained, and the quality of the rotating machine is determined as a function of the distance from the feature quantity distribution line. Define the probability density function of the index used for judgment (failure probability), and when the distance between the feature quantity and the feature quantity distribution line becomes a certain value, the judgment index in the probability density function takes a specific value Determine the value of the standard deviation and store this value in your computer's memory,
When a feature amount is obtained when diagnosing a rotating machine, a distance between the feature amount and the feature amount distribution line is obtained, and a determination index is obtained by using the distance value and the standard deviation value obtained in advance. The gist is to diagnose the quality of the rotating machine during operation by calculating the failure probability and setting a threshold for the value.

  Whether or not there is an abnormality occurring in the rotating machine that is operating can be diagnosed without stopping the rotating machine even when the magnitude of the load changes variously with the load connected.

Moreover, the quality check system for the rotating machine of the present invention is:
The current flowing in each phase of a normal rotating machine operating with a load connected is measured for each of the various loads, and the specific current included in the rotating machine is measured for various loads. The frequency component current is extracted by performing frequency analysis and the like, the feature amount is obtained from the current of the specific frequency component, the region where the feature amount is distributed is obtained, and the region where the feature amount is distributed in the normal rotating machine A reference line (feature distribution line) is obtained, and a probability density function of an index (failure probability) used for determining the quality of a rotating machine is defined as a function of the distance from the feature distribution line. When the distance from the distribution line becomes a certain value, the standard deviation value in the probability density function is determined so that the determination index takes a specific value, and this value is stored in the memory of the computer,
When a feature amount is obtained when diagnosing a rotating machine, a distance between the feature amount and the feature amount distribution line is obtained, and a determination index is obtained by using the distance value and the standard deviation value obtained in advance. The gist is to diagnose the quality of the rotating machine during operation by calculating the failure probability and setting a threshold for the value.
In this way, there is a system that can diagnose the presence or absence of an abnormality that occurs in a rotating machine in operation without stopping the rotating machine even when the load changes in various ways with the load connected. It can be easily constructed.

In the rotating machine pass / fail diagnosis system of the present invention, the distribution area of the feature amount when the rotating machine is normal is not approximated by a single feature amount distribution line, but is approximated by a plurality of lines divided into sections. Can also be diagnosed.
In this way, the distribution area can be approximated more accurately than when the feature quantity distribution area is approximated by a single line.

Further, in the rotating machine pass / fail diagnosis system of the present invention, it is also possible to obtain 100 or more feature quantities to be obtained for the loads of various sizes.
By doing so, the error in obtaining the distribution line is reduced.

Further, in the rotating machine pass / fail diagnosis system of the present invention, when extracting the feature amount, a filter may be used to remove a noise component included in the current.
In this way, it is possible to extract feature quantities that are not affected by noise, to obtain feature values with correct values, and to obtain accurate values for the obtained feature quantity distribution lines.

In the rotating machine quality diagnosis system according to the present invention, the feature quantity is indicated in a three-dimensional space, and the function quantity from various angles can be obtained by providing a function of rotating the three-dimensional spatial distribution on a computer screen. It can also be configured to check the distribution region and visually check the change in the feature amount when the rotating device is normal and abnormal.
In this way, by rotating the feature distribution displayed in the three-dimensional space and viewing the state of the distribution from various angles, it is possible to visually check the change of the feature when the rotating device is normal and abnormal Will be able to.
As will be described later, the number of feature amount distribution lines can be used to understand how to approximate the feature amount distribution.

Moreover, in the quality determination system for a rotating machine according to the present invention, the reference line for the region in which the feature amount is distributed in the normal rotating machine may be a straight line or a curved line.
In this way, an optimal distribution can be selected.

  It is a system outline figure showing the whole flow of this diagnostic system.   It is a feature-value distribution map obtained from a normal electric motor.   It is a feature-value distribution map obtained from the electric motor at the time of short circuit occurrence.   It is a histogram figure of distance d of the feature-value at the time of normal, and a feature-value distribution line.   It is a histogram figure of distance d of the feature-value at the time of a short circuit, and a feature-value distribution line.   It is a figure which shows the probability density function of a failure probability.   It is a figure which shows the probability density function in the case of a linear function.   It is a figure which shows the probability density function in the case of an exponential function.   It is explanatory drawing which extracts the maximum value of the positive side of the electric current which flows into each phase as a feature-value.   It is a figure which shows the change of the maximum value Iwmax of the positive side of the load current which flows into a W phase when changing load.   It is a figure which shows the failure probability calculated when changing the load of FIG.   It is a figure which shows the change of the maximum value Iwmax of the positive side of the load current at the time of generating a short circuit.   It is a figure which shows the failure probability calculated when changing the load of FIG.   It is explanatory drawing of how to obtain | require a feature-value distribution line.   It is a figure of feature-value distribution and a feature-value distribution line.   It is a figure which shows the distance of each feature-value and a feature-value distribution line.   It is a figure which shows the effect of a noise removal filter.   It is a figure which shows the feature-value extraction from the electric current containing many frequency components.

Next, the form for implementing this invention is demonstrated.
The overall flow of this diagnostic system is shown in FIG.
The processing up to diagnosis in this diagnosis system can be roughly divided into two processes. One is a pre-processing process (pre-processing step) corresponding to the previous stage of diagnosis, and the other is a diagnostic process (diagnosis) that actually calculates the failure probability and determines the presence or absence of a short circuit that occurs in the winding inside the rotating machine. Process).

1. Pretreatment process (pretreatment process)
In the first pretreatment process, the characteristic amount, that is, the maximum value on the positive side of the current flowing in each phase is measured with a normal rotating machine in operation.
A feature amount distribution line is derived from the three-dimensional feature amount distribution. Then, a distance d between the derived feature quantity distribution line and each feature quantity is calculated, and a histogram thereof is obtained.
Based on the histogram, a standard deviation σ of a probability density function for determining a failure probability is obtained and stored in the storage unit.
In this state, the process proceeds to the next diagnosis process.

2. Diagnosis process (diagnosis process)
In the diagnosis process, the characteristic amount, that is, the maximum value on the positive side of the current flowing in each phase is measured in a state where the rotating machine is operated. The feature amount of time and wherein the amount F _d.
Next, a distance d between the feature amount _Fd and the feature amount distribution line is obtained. The feature amount distribution line used here is obtained in the preprocessing process.
Now, it is assumed that d = d ₁ is obtained as the distance d between the feature amount F _d and the feature amount distribution line. The failure probability P can be calculated by using d = d ₁ and the later-described equation (2).

  By setting a threshold value etc. for the value of the failure probability thus obtained, it is determined whether the stator winding of the rotating machine operating while the load fluctuates is normal or whether there is a failure such as a short circuit. Can be done statistically.

Here, in describing a specific short-circuit diagnosis method, a theoretical explanation will be given by taking an electric motor as an example of a rotating machine. As an example of the failure, consider a case where a short circuit occurs in the coil of the stator winding inside the electric motor.
Now, connect the power supply to a normal motor. A load is connected to the electric motor. The current flowing in each phase of the electric motor that is operating while changing the magnitude of the load is measured synchronously, and the feature value is extracted therefrom.
Here, as shown in FIG. 9, the maximum value on the positive side of the current flowing in each phase of the U phase, the V phase, and the W phase in a time interval in which each phase current waveform includes at least one cycle is included as a feature amount. If it is a value, the feature amount is (Iumax, Ivmax, Iwmax).
A feature amount obtained when the load has a certain value (for example, load state 1) is defined as a feature amount F ₁ (Iumax 1, Ivmax 1, Iwmax 1).

Next, a feature amount obtained when the load takes another value (for example, load state 2) is defined as a feature amount F ₂ (Iumax 2, Ivmax 2, Iwmax 2).
When this is similarly obtained for various load states, a large number (N points) of feature values F _N (Iumax N, Ivmax N, Iwmax N) are obtained.
In this way, more feature quantities than 100 points are obtained.
When obtaining a feature quantity distribution line, a large number of feature quantity points are required. If one of the many feature points is distributed in a region that is greatly deviated due to the influence of noise, etc., if the number of feature points is large, the effect of this point will be affected when the feature amount distribution line is obtained. However, if the number of feature points is small, it is greatly affected by this point. For this reason, it is impossible to correctly obtain a feature amount distribution line that approximates a feature amount distribution region. From the above, 100 or more feature points are required.
This time, when 800 feature values are measured and displayed in a feature distribution in a three-dimensional space, the result is as shown in FIG.
According to FIG. 2, although the magnitude | size of load changes variously, it turns out that the feature-value obtained at that time is distributed substantially linearly.

Next, in the internal stator winding of the electric motor, with the adjacent coils inside being short-circuited, the electric motor is continuously operated while changing the load value in the same manner as described above, and the feature amount is obtained. Then, feature values of 100 points or more are obtained.
This time, 800 feature values are obtained and plotted in a three-dimensional space together with the feature distribution when the windings are normal as shown in FIG. 2, and this is newly shown in FIG.
According to the result of FIG. 3, when a short circuit occurs inside the motor, the characteristic amount obtained in the normal state and the distribution area are different, so this information can be used for the diagnosis of the motor.

  In this way, as a result of measuring the current flowing through each phase of the rotating machine operating with the load connected for each load of various sizes, the distribution of the characteristic amount obtained from the current flowing through the rotating machine is almost linear. The present invention has been devised based on the fact that the current distribution region at the normal time and the current distribution region at the occurrence of a short circuit are different.

Next, the pretreatment process for implementing the short-circuit diagnosis method will be specifically described.
First, in the feature quantity distribution obtained from the normal motor shown in FIG. 2, a straight line is obtained as a line approximating the distribution. This straight line is a straight line passing through the middle of the distribution. An image diagram is shown in FIG.

In FIG. 2, when the value of the feature value is small, that is, when the motor is operating near no load, the slope of the straight line approximating the feature value distribution is slightly smaller than in other load states. Is different.
Therefore, instead of obtaining one feature amount distribution line from the feature amount distribution of FIG. 2, the range in which Iwmax is distributed (here, between 6A and 14A) is divided at appropriate intervals, and the feature amount distribution is divided for each section. Let's find a straight line.
By doing so, for example, as shown in FIG. 15, even when the distribution region of the feature amount is curved, the distribution can be approximated by several straight lines, and the feature amount distribution region can be approximated by one straight line. This makes it possible to approximate the distribution region more accurately than the case. The several normal distributions of feature quantities obtained in this way are referred to as “feature distribution lines” for convenience.

For example, the feature amount distribution line is as follows.
If the current Iw flowing in one phase, here the W phase, is used as a reference, the value of the current feature value Iwmax obtained from Iw is between 6A and 14A, so this interval is divided into 2A intervals, and Iwmax is 6 to Feature amount distribution line at 8A is straight line 1, Feature amount distribution line at 8-10A is straight line 2, Feature amount distribution line at 10-12A is straight line 3, Feature amount distribution line at 12-14A Is a straight line 4.

Then, according to the value of Iwmax, as shown in FIG. 16, a distance d when a perpendicular perpendicular to the corresponding feature quantity distribution line is drawn from each feature quantity is obtained.
For example, if a feature value having an Iwmax value of 7.0 A is given, the above-described straight line 1 is used as the feature quantity distribution line at that time, and the distance d between the feature quantity point and the straight line 1 is obtained. .

Thus, the distance d between each feature quantity and the feature quantity distribution line is regarded as a new feature quantity.
By doing so, the three-dimensional feature values (Iumax, Ixmax, Iwmax) can be set as the one-dimensional feature value d, and the handling becomes easy.

FIG. 4 and FIG. 5 show the histograms of the distance d between each feature point at normal time and short circuit and the feature amount distribution line obtained above. 4 and 5, the horizontal axis represents the distance d, and the vertical axis represents the occurrence frequency, that is, the number of occurrences.
In the feature amount distribution of FIG. 3, since there are few regions where the feature amounts overlap between the normal winding and the short circuit, the overlap is also reduced on the histogram.
A parameter (in this case, standard deviation σ) necessary for diagnosis is calculated from this histogram.

ここでσは標準偏差を表す。また、（１）式の確率密度関数において、平均は特徴量分布直線上に相当するｄ＝０とする。さらに、距離ｄは常に正の値を取るため、ここでは一般的な正規分布の式を２倍した式を用いる。こうすることで、ｆ（ｄ）を距離ｄ＝０から∞まで積分した値が１となる。Next, a diagnostic process for carrying out the short-circuit diagnostic method will be specifically described.
This time, we define the failure probability as an index to evaluate the state of the winding.
The probability density function f (d) of this failure probability is expressed by the following normal distribution expression as a function of the distance d.

Here, σ represents a standard deviation. In the probability density function of the equation (1), the average is d = 0 corresponding to the feature amount distribution line. Furthermore, since the distance d always takes a positive value, a formula obtained by doubling a general normal distribution formula is used here. By doing so, the value obtained by integrating f (d) from the distance d = 0 to ∞ is 1.

The characteristic amount, that is, the maximum value on the positive side of the current flowing in each phase is measured in a state where the rotating machine is operated. The feature amount at this time is defined as a feature amount Fd (Iumax d, Ivmax d, Iwmax d). Assuming that d = d ₁ is obtained as the value of the distance d between the feature quantity Fd and the feature quantity distribution line, the failure probability P at this time can be calculated by the following equation.

The failure probability P takes various values depending on how the standard deviation σ is determined.
In the present electric motor, as shown in FIG. 4, the maximum frequency is obtained when the distance d when the winding is normal is d = 0.06. Therefore, the standard deviation is determined so that the failure probability at d = 0.06 is 20%, that is, P = 0.20.
This means that the probability that a feature amount exists in a space where the distance from the feature amount distribution line is d = 0.06 or less is 20%.
The standard deviation σ determined in this way is 0.244. Save this value.

FIG. 6 shows the probability density function f (d) when the standard deviation σ is 0.244.
This time, the probability density function f (d) of the failure probability is expressed by the expression expressed by the expression (1), but this may be expressed as a form of a linear function or an exponential function.
For example, in the case of a linear function, equation (3) is obtained. Here, A and B are positive constants.
In the case of an exponential function, the form is as in equation (4). Here, A and B are positive constants.
The constants A and B are determined depending on how the value of the failure probability P is determined at the distance d.

The waveform in the case of a linear function is as shown in FIG. 7, and in the case of an exponential function, it is as shown in FIG.
f (d) = − Ad + B (3)
f (d) = Aexp (−Bd) (4)

  Below, in order to confirm that the presence or absence of the short circuit which generate | occur | produces with an internal stator winding of an electric motor can be determined, a diagnostic result is demonstrated with experimental data.

First, a diagnosis is performed on an electric motor in which the stator winding is in a normal state.
Current measurement for each phase is performed at 30-second intervals in synchronization. In one measurement, for example, as shown in FIG. 9, the maximum value on the positive side of the current of each phase is derived in a time interval in which each phase current waveform includes at least one cycle. The feature value. A failure probability is calculated and evaluated for each of 60 points of data obtained when the magnitude of the load connected to the motor is arbitrarily changed.

10 and 11 show the maximum value Iwmax on the positive side of the load current flowing in the W phase when the load is varied, and the change in the failure probability calculated at this time, respectively.
In FIG. 10, the horizontal axis represents the number of data, and the vertical axis represents the maximum value Iwmax on the positive side of the current. Further, the horizontal axis in FIG. 11 is the number of data, and the vertical axis is the failure probability calculated for the feature amount obtained at that time.
The feature value Iwmax varies depending on the load, and the obtained failure probability varies, but the value is generally low. The average failure probability for 60 points of data is 23.5%.

Next, the same evaluation is performed when a short circuit is generated in the stator winding.
The maximum value on the positive side of the load current at this time and the failure probability calculated at that time are shown in FIGS. 12 and 13, respectively.
When a short circuit occurs in the stator winding, the obtained feature quantity distribution is shifted from the normal time, so that the value of the distance d increases, and as a result, the failure probability changes at a high value. I understand. The average failure probability for these 60 feature quantities is 91.9%.
From the above results, even when the size of the load connected to the motor changes variously, it is possible to determine the presence or absence of a short circuit occurring in the motor's internal stator winding, regardless of the size of the load. Can be confirmed.

This time, the maximum value on the positive side of the current flowing in each phase was used as the feature value. However, in general, a noise component is included in the current flowing through the rotating machine. Therefore, if the noise component is not removed from the current, the obtained feature amount is also affected by the noise, and the variation of the distribution region of the feature amount shown in FIGS. 2 and 3 becomes large. Therefore, a low-pass filter LPF and a moving average filter are used to remove noise components included in the current flowing through the rotating machine.
As shown in FIG. 17, for example, even when noise is superimposed on the current of the fundamental wave component of 60 [Hz], a clean fundamental wave component can be extracted by using a low-pass filter and affected by the noise. Features that are not available can be extracted.
The cut-off frequency of the LPF is 20 times or less of the fundamental frequency component of the current. For example, if the fundamental wave component of the current is 60 [Hz], the cutoff frequency is 1200 [Hz] or less.
By removing extra noise components in this way, it is possible to obtain a feature value having a correct value, and an accurate value can also be obtained from the obtained feature value distribution line.

When various frequency components are included in the current, each frequency component can be obtained by performing frequency analysis such as FFT.
As shown in FIG. 18, each frequency component included in the current can be calculated by performing frequency analysis. Then, paying attention to a specific frequency component (for example, 60 [Hz]), the spectral intensity S ₆₀ may be used as a feature amount.
Further, as described above, the maximum value on the positive side in the current of the specific frequency component may be used as the feature amount.
For example, a method of extracting a specific frequency component when the current includes various frequency components will be described with reference to FIG. Here, the specific frequency component is 60 Hz.
First, FFT (Fast Fourier Transform) is performed on a current including various frequency components. At this time, a frequency spectrum for this current is obtained.
Only the value of the spectral component corresponding to 60 Hz is left, and all other spectral components are set to zero.
Next, IFFT (Inverse Fast Fourier Transform) is performed on the frequency spectrum that leaves only the 60 Hz component. By doing so, only the current waveform of the 60 Hz component can be extracted. The maximum value on the positive side in the current of the specific frequency component thus obtained may be used as the feature amount.
Then, these frequency analyzes may be performed for all three-phase currents to obtain the distribution of the extracted feature values.

  In the present embodiment, a short circuit in the stator winding of the motor was considered as an example of the failure, but in addition to this, when the bearing inner ring, outer ring, rolling element, etc. are damaged, the presence or absence of damage in the rotor is checked. It may be a target of diagnosis.

  In the above-described embodiment, the reference of the region in which the feature amount is distributed in a normal rotating machine is illustrated by a straight line, but a curved line is not limited to a straight line.

Claims (7)

The current flowing in each phase of a normal rotating machine operating with a load connected is measured for each load of various magnitudes, and feature quantities are obtained for various load magnitudes. The area in which the quantity is distributed is obtained, the reference line (feature quantity distribution line) of the area in which the feature quantity is distributed in the normal rotating machine is obtained, and the quality of the rotating machine is determined as a function of the distance from the feature quantity distribution line. Define the probability density function of the index used for judgment (failure probability), and when the distance between the feature quantity and the feature quantity distribution line becomes a certain value, the judgment index in the probability density function takes a specific value Determine the value of the standard deviation and store this value in your computer's memory,
When a feature amount is obtained when diagnosing a rotating machine, a distance between the feature amount and the feature amount distribution line is obtained, and a determination index is obtained by using the distance value and the standard deviation value obtained in advance. A rotating machine pass / fail diagnostic system characterized in that the failure probability of the rotating machine is calculated and a threshold value is provided to diagnose the failure of the rotating machine during operation. The current flowing in each phase of a normal rotating machine operating with a load connected is measured for each load of various sizes and included in the current flowing in the rotating machine for loads of various sizes. The current of the specific frequency component is extracted by performing frequency analysis, etc., the feature amount is obtained from the current of the specific frequency component, the region in which those feature amounts are distributed is obtained, and the distribution of the feature amount in the normal rotating machine A reference line (feature distribution line) for a region to be determined, and a probability density function of an index (failure probability) used to determine the quality of a rotating machine as a function of the distance from the feature distribution line is defined The standard deviation value in the probability density function is determined so that the judgment index takes a specific value when the distance between and the feature amount distribution line becomes a certain value, and this value is stored in the memory of the computer. When the diagnosis of the rotating machine, When obtained, the distance between the feature quantity and the feature quantity distribution line is obtained, and the failure probability as the determination index is calculated using the distance value and the standard deviation value obtained in advance. A rotating machine pass / fail diagnostic system characterized by diagnosing the quality of a rotating machine during operation, for example, by setting a threshold value.   2. The diagnosis is performed by approximating the distribution area of the feature quantity when the rotating machine is normal with a plurality of lines instead of approximating with one feature quantity distribution line. Item 3. A diagnostic system for rotating machines according to item 2.   The system for diagnosing pass / fail of a rotating machine according to any one of claims 1 to 3, wherein 100 or more feature quantities are obtained as the number of feature quantities obtained for the loads of various sizes.   The system for diagnosing pass / fail of a rotating machine according to any one of claims 1 to 4, wherein a filter is used to remove a noise component included in the current when extracting the feature amount.   By displaying the feature amount in a three-dimensional space and having the function of rotating the three-dimensional space distribution on the computer screen, the distribution region of the feature amount from various angles can be confirmed, and the rotating device can be operated normally and abnormally. The system for diagnosing pass / fail of a rotating machine according to any one of claims 1 to 5, characterized in that a change in a feature amount with time is visually confirmed.   The system for diagnosing pass / fail of a rotating machine according to any one of claims 1 to 6, wherein a line serving as a reference of a region in which the feature amount is distributed in the normal rotating machine is a straight line or a curved line.

Images