JPH063215A - Diagnostic method and diagnosing device for rotary equipment - Google Patents

Abstract

PURPOSE:To diagnose abnormality generation of a rotary equipment with high sensitivity and high accuracy. CONSTITUTION:A time-series data sampled at a normal time is applied to an autoregressive moving average model by an adaptive digital filter, and its parameters Ak and Bk are calculated (#1). At diagnosis, a forecast value Y (j) is calculated (#3) with the autoregressive moving average model using a time-series data X (j) similarly sampled and X (j) and its error E (j) are acquired (#4). White color audit or the like (#5) is carried out by this E (j).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a diagnostic method and diagnostic apparatus for rotating equipment such as electric motors and turbines.

[0002]

2. Description of the Related Art In rotating equipment, wear of contact parts such as bearings is an unavoidable phenomenon, and as wear continues, abnormal noise is generated, or an electric current in an electric motor fluctuates abnormally.
Eventually, the operation itself becomes impossible. Japanese Unexamined Patent Publication (Kokai) No. 2-8893 discloses a diagnostic device for checking the deterioration state of such equipment.
The one described in Japanese Patent Publication No. 9 is known. This device detects the vibration signal generated by rotating equipment and applies it to the autoregressive model as time series data to obtain the coefficient of the time series data as the autoregressive parameter, while determining the frequency component as the power spectral density and the autoregressive parameter. The diagnosis is performed by substituting the value and the power spectrum density data into a predetermined evaluation function to calculate the evaluation value and comparing it with the evaluation value obtained in advance in the same manner in the normal state.

[0003]

However, such a conventional device has the following problems. (1) The autoregressive model expresses the contribution rate of past data to the present time as an autoregressive parameter. Therefore, the influence of periodic data is expressed as the magnitude of the parameter coefficient. In the above-mentioned device, the evaluation is performed by comparing the parameter value at the time of normal and the parameter value at the time of diagnosis. Therefore, the variation of the value only describes a part of the state change, and it cannot be said that it is unreliable to state the state change only by the change. (2) Similarly in frequency analysis, evaluation is performed by comparing the spectral density detected during normal operation with the spectral density detected during diagnosis. Even in this case, only the level change as an aspect of the state change is observed, and sufficient reliability cannot be obtained by the state determination based only on the change. (3) It is desired that the level of the signal used for diagnosis is sufficiently small in the normal state and that the amount of change in the signal level when the state changes is large, that is, the sensitivity to the state of the detected signal is sufficiently high. However, since the signal is present even in the normal state, the signal itself in the normal state masks the detected signal and makes it difficult to detect the change in the state.

The present invention has been made to solve the above-mentioned problems of the prior art. The time-series data is applied to an autoregressive moving average model by an adaptive digital filter, its parameters are determined, and this is used. Therefore, it is an object of the present invention to provide a method and apparatus capable of high reliability, high accuracy, and high speed diagnosis.

[0005]

According to a first aspect of the present invention, there is provided a method for diagnosing a rotating device, comprising: a method for diagnosing a state of a rotating device based on time series data of signals generated while the rotating device is rotating. The time series data of is fitted to an autoregressive moving average model by an adaptive digital filter,
The parameters of the autoregressive moving average model are calculated, and at the time of diagnosis, the error of the difference between the actual signal and the predicted signal by the autoregressive moving average model having the parameters is calculated, and the state of the rotating equipment is diagnosed based on this error. It is characterized by

According to a second aspect of the present invention, there is provided a diagnostic apparatus for rotating equipment, which is an apparatus for diagnosing a state of a rotating equipment based on time series data of a signal generated while the rotating equipment is rotating. And an adaptive digital filter for calculating the parameters of the autoregressive moving average model to which the time-series data should be applied,
Means for predicting and calculating time series data with an autoregressive moving average model having the calculated parameters, and means for calculating an error in the difference between the predicted time series data and the actual time series data, based on the error It is characterized in that it is designed to diagnose the state of the rotating equipment.

[0007]

In the present invention, first, the optimum autoregressive moving average parameter for the state obtained from the rotating device in the reference state, that is, the normal state, and the detection signal (for example, vibration signal) is an adaptive digital filter (hereinafter referred to as ADF). And an autoregressive moving average model is created. Generally, the ADF that expresses the autoregressive moving average model is expressed as follows.

[0008]

[Equation 1]

Here, X (j) is the input data at time j, Y (j) is the predicted value by ADF, and A (k, j) and B (k, j) are the one step as the kth model parameter at time j. What made the prediction. Under these conditions, parameters A (1,2 ... n) and B (1,2 ... m) that minimize E (j) are obtained.

The algorithm for minimizing E (j) is based on the steepest gradient method, and ADF is constructed as a recursive-ADF, and White's algorithm or Feintu modified it.
It can be realized by the ch algorithm and the probability approximation algorithm. After preparing like this,
At the time of diagnosis, the state detection target signal is collected from the rotating device.
On the other hand, the signal reproduced by the model parameter determined by the ADF is predicted, and the prediction error signal obtained by the difference from the predicted signal is calculated. As can be seen from the above equation, the following calculation is performed.

[0011]

[Equation 2]

A whiteness test is performed on the prediction error thus obtained. This includes basic statistics (mean value, variance) and frequency distribution (autocorrelation function calculation, FFT calculation)
Is calculated. If the parameters are determined accurately, the prediction error is basically equal to white noise. Therefore, whether or not the state has changed can be detected by performing a whiteness test of the prediction error.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings showing its embodiments. FIG. 1 is a schematic diagram showing an electric motor to be diagnosed and a device of the present invention. Bearing 4, which supports the driven shaft connected by the electric motor 1 and its output shaft by the coupling 2.
5 is fixed on a base 6, and a load 3 is connected between both bearings 4 and 5. An acceleration sensor 7 is mounted on the outer surface of the bearing 5 to detect vibration of the electric motor. The output of the acceleration sensor 7 is input to the amplifier 8, the amplified output is input to the A / D converter 9 and quantized therein, and the quantized data is temporarily stored in the storage device 10. The data in the memory device 10 is sequentially read by the CPU 11 and the CP
U 11 performs the arithmetic processing described later to display the arithmetic progress and the diagnostic result on the display device 13 and, if necessary, the communication device.
The diagnostic result is reported to another device (not shown) via 12.

FIG. 2 is a flowchart showing the processing procedure of the present invention. ADF is realized by CPU 11. First, when the motor 1 can be determined to be normal by another means, that is, in the reference state, the parameters Ak and Bk of the autoregressive moving average model are calculated (# 1). This is E (j)
Is calculated by a known algorithm that minimizes.
The parameters thus calculated are temporarily stored in the storage device 10.

Therefore, in the actual diagnosis, the following step # 2
Perform ~ # 8. That is, the data X (j) sequentially stored in the memory device 10 at the time of diagnosis is sequentially read (# 2), and Y (j) is calculated by the equation (1) (# 3). Next, the difference between the predicted value Y (j) thus obtained and the measured value X (j) is calculated by the equation (2) (# 4).

The operation result is evaluated using E (j) thus obtained. First, perform whiteness test (#
5) If the color is white, it ends as normal. In this determination, for example, the normalized variance is obtained using the correlation function of E (j), and it is verified whether or not this is within the range of ± 1 / √N (N: number of data). If it is colored, check the variation factor (# 6). This means that a remarkably rapid change is a state change due to an abnormality of the electric motor, while a gradual change is an external factor change. As a method thereof, a method using an average value, a variance value, etc. of E (j) can be mentioned. For example, when the average value is used, the average value is calculated, and if the average value that is originally 0 or close to 0 deviates from a predetermined value by a large amount, it is determined that there is a change in an external factor. For example, an increase in background noise can be cited as an external factor. In such a case, the parameters Ak and Bk are recalculated (# 7), and subsequent diagnosis is performed using the new parameter.

On the other hand, when it is determined that an abnormality has occurred, the device state is determined (# 8). For this purpose, for example, autocorrelation is calculated with respect to the prediction error E (j) to detect periodic pulse noise. The state of the equipment is detected by comparing this pulse noise with self-cases and empirical rules. If it is desired to detect aged deterioration of rotating equipment as an abnormality, the parameter recalculation or change in step # 7 need not be performed.

FIG. 3 shows an example of normal data. The horizontal axis indicates the data number j, and the vertical axes indicate X (j), Y (j), and E, respectively.
The standardized value of (j) is shown. 16 revolutions per motor
Corresponds to 02 data. E (j) is almost 0. FIG. 4 shows a similar signal when an abnormality occurs in the bearing. The prediction error signal E (j) periodically shows a large value. From this, it is clear that the prediction error E (j) is useful for determining the occurrence of abnormality.

Next, the effect of ADF will be described. Figure 5
(a) and (b) show the results of the autocorrelation calculation under normal conditions, without ADF (a) and with ADF (b), respectively.
(a) and (b) show the results of autocorrelation calculation when a bearing failure occurs in ADF.
It is shown without (a) and with ADF (b). 5 and 6, the vertical axis represents the normalized autocorrelation coefficient, and the horizontal axis represents the time converted from the data collection time and the data number. Level comparison can be detected without ADF by comparing Fig. 5 (a) and Fig. 6 (a). However, it is difficult to point out the characteristic frequency component F ₁ that appears remarkably when a bearing abnormality occurs. On the other hand, according to the comparison between FIG. 5 (b) and FIG. 6 (b), in the case of ADF processing, this frequency component can be easily detected, and it is possible to estimate the fault occurrence or the faulty part or site.
As shown in Fig. 6 (b), it is AD that the characteristic frequency component appears remarkably.
This is because it effectively suppresses the periodic component that appears even when F is normal.

[0020]

As described above in detail, in the case of the present invention,
ADF can suppress normal signals and extract only characteristic signals, enabling higher-sensitivity and higher-precision diagnosis than before. Also
Since the ADF calculates the parameters of the autoregressive moving average model, the parameters can be easily determined even when the diagnostic device is applied to various diagnostic target devices. Further, according to the present invention, the presence or absence of fluctuation is detected based on the magnitude of the prediction error signal instead of comparing the parameters at the time of normal and at the time of diagnosis, so that highly reliable diagnosis can be performed unlike the prior art. Furthermore, in the present invention, since ADF is used for the calculation of the autoregressive moving average model and the determination of its parameters, it is possible to perform high-speed calculation as compared with the case where it is difficult to form a digital filter such as the recursive least squares method. There is.

[Brief description of drawings]

FIG. 1 is a schematic view of a device of the present invention.

FIG. 2 is a flowchart showing a processing procedure of the present invention.

FIG. 3 is a data chart under normal conditions.

FIG. 4 is a data chart at the time of abnormality.

FIG. 5 is a chart showing the result of autocorrelation calculation under normal conditions.

FIG. 6 is a chart showing an autocorrelation calculation result at the time of abnormality.

[Explanation of symbols]

 7 Acceleration sensor 9 A / D converter 10 Storage device 11 CPU

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] July 17, 1992

[Procedure Amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction content]

On the other hand, when it is determined that an abnormality has occurred, the device state is determined (# 8). For this purpose, for example, autocorrelation is calculated with respect to the prediction error E (j) to detect periodic pulse noise. The state of the equipment is detected by comparing this pulse noise with accident cases and empirical rules. If it is desired to detect aged deterioration of rotating equipment as an abnormality, the parameter recalculation or change in step # 7 need not be performed.

[Procedure Amendment 2]

[Document name to be corrected] Drawing

[Name of item to be corrected] Figure 5

[Correction method] Change

[Correction content]

[Figure 5]

[Procedure 3]

[Document name to be corrected] Drawing

[Name of item to be corrected] Figure 6

[Correction method] Change

[Correction content]

[Figure 6]

Claims (2)

[Claims] 1. A method for diagnosing the state of a rotating device based on time-series data of a signal generated during rotation of the rotating device, wherein the time-series data in a reference state is converted into an autoregressive moving average model by an adaptive digital filter. Fit,
The parameters of the autoregressive moving average model are calculated, and at the time of diagnosis, the error of the difference between the actual signal and the predicted signal by the autoregressive moving average model having the parameters is calculated, and the state of the rotating equipment is diagnosed based on this error. A method for diagnosing rotating equipment, which is characterized in that 2. An apparatus for diagnosing the state of a rotating machine based on time series data of a signal generated during rotation of the rotating machine, wherein means for quantizing the signal to obtain time series data and time series data are provided. The adaptive digital filter that calculates the parameters of the autoregressive moving average model to which this should be applied, the means for predicting and calculating the time series data with the autoregressive moving average model having the calculated parameters, and the predicted time series data and the actual And a means for calculating an error of a difference with respect to the time series data, and the state of the rotating machine is diagnosed based on the error.