JP2006113002A - Abnormality diagnosis system for mechanical equipment - Google Patents

Abstract

To provide an abnormality diagnosis system capable of performing an abnormality diagnosis with high accuracy without erroneously detecting a noise signal as an abnormal signal even under a condition where the S / N ratio between the abnormal signal and the noise signal is small.
An envelope processing unit 3 for obtaining an envelope of a detection signal, an FFT unit 4 for converting the envelope into a frequency spectrum, smoothing by moving and averaging the frequency spectrum, and further smoothing and differentiating the spectrum to obtain a differential coefficient. A peak detection unit 5 that detects a frequency point where the sign of the signal changes from positive to negative as a peak, extracts those that exceed a predetermined threshold, sorts them, and detects a higher one as a peak, and a peak detection unit 5 And a diagnosis unit T for diagnosing an abnormality based on the peak.
[Selection] Figure 1

Description

  The present invention relates to an abnormality diagnosis technique for mechanical equipment including a bearing such as a railway vehicle, an aeronautical machine, a wind power generator, a machine tool, an automobile, an iron making machine, a papermaking machine, a rotating machine, and the like. The present invention relates to an abnormality diagnosis technique for mechanical equipment that diagnoses abnormality of a bearing or a bearing-related member in the mechanical equipment by analyzing generated sound or vibration.

  Conventionally, as this type of abnormality diagnosis technology, a signal representing sound or vibration from a sliding member of a mechanical equipment or a sliding member-related member is detected, and a frequency spectrum of the detected signal or its envelope signal is obtained, and the frequency spectrum is obtained. From this, only the frequency component due to the abnormality of the sliding member of the mechanical equipment or the sliding member related member of the mechanical equipment is extracted, and the abnormality in the sliding member used in the mechanical equipment is extracted according to the size of the extracted frequency component. What diagnoses the presence or absence of this is known (refer patent document 1).

  Further, sound or vibration generated from the rotating body or the rotating body-related member is detected, a signal in a frequency band necessary for diagnosis is extracted from the detected signal, an envelope (envelope) of the extracted signal is obtained, and the obtained envelope The frequency analysis is performed to obtain the magnitude of the fundamental frequency component of the frequency caused by the abnormality of the rotating body or the rotor-related member and the magnitude of the frequency component that is a natural number multiple of the frequency component. It is also known that the magnitude of the frequency component that is a multiple of the natural number is compared, and at least the comparison result is used as a criterion for judging the abnormality of the mechanical equipment (see Patent Document 2).

  In addition, analog signals of sound or vibration generated from mechanical equipment are converted into digital signals by A / D (analog / digital) conversion to generate measured digital data, and frequency analysis and envelope analysis are performed on this measured digital data. Measured frequency spectrum data is generated by performing appropriate analysis processing, and abnormalities in the mechanical equipment are determined by the presence or absence of peaks in the measured frequency spectrum data for the primary, secondary, and quadratic values of the frequency components caused by abnormalities in the mechanical equipment. There is also known one that diagnoses the presence or absence of (see Patent Document 3).

  In addition, the envelope waveform of vibration acceleration is converted into a digital signal, and the vibration spectrum distribution of the digitized vibration data for each time is obtained, and the rotation speed of the rolling bearing at the time of vibration measurement is obtained every moment, and the time change of the rotation speed is obtained. The time variation pattern of the frequency of the peak spectrum in the vibration spectrum distribution coincides with that of the vibration spectrum distribution. Further, the frequency of the peak spectrum at an arbitrary time is determined based on the rolling bearing rotation speed and the rolling bearing geometric dimension. It is also known that it is determined that damage has occurred in a specific part of a rolling bearing when it matches the characteristic frequency (see Patent Document 4).

Although these patent documents do not specify a method for detecting a peak of a frequency indicating an abnormality, if an abnormality such as a peeling life of a bearing or an eccentricity of a rotating shaft of a machine occurs, a signal indicating the abnormality (abnormal signal) ) Frequency peak can be easily obtained by the integrated average of the frequency spectrum. Accumulated average is a technique often used in the field of frequency analysis such as fast Fourier transform (FFT) analysis because it is effective in removing random noise.
JP 2003-202276 A JP 2003-232674 A JP 2003-130763 A Japanese Patent Laid-Open No. 09-113416

  However, vibration and acoustic sensors are subject to external impact sounds and friction sounds, and in the case of moving objects, acceleration due to turning acts, so abnormalities may be erroneously detected due to these unsteady disturbances. Many. For this reason, the frequency peak detection method based on the integrated average may not be effective because it tends to be affected by a change in speed, an external impact sound, or the like if the number of integrations is increased.

  In the case of an abnormality due to a small scratch, peeling, rust, etc. before reaching the end of its life, the signal power from the vibration sensor or acoustic sensor is often small enough to be buried in mechanical noise or electrical noise. For this reason, it is often impossible to use a method in which a threshold is set and only a signal having a power higher than that value is extracted at an abnormal prediction stage before the lifetime. The most troublesome problem in predicting an abnormality is that the noise signal is converted into an abnormal signal when the S / N ratio between the abnormal signal or the signal indicating the abnormal sign (abnormal predictive signal) and the noise signal is small. Or, it is erroneously determined as an abnormal sign signal. It is advantageous to improve the accuracy of abnormal prediction of bearings, etc., so as not to overlook extremely small abnormal signals and abnormal predictor signals, but as a result, noise signals are mistakenly determined as abnormal signals and abnormal predictor signals. As a result, the machine equipment is frequently stopped and inspected, resulting in an increase in operating costs.

  The present invention has been made in view of the above-described circumstances, and an object of the present invention is to make a noise signal abnormal or abnormal sign signal even under a condition where the S / N ratio between the abnormal signal or abnormal sign signal and the noise signal is small. It is an object of the present invention to provide an abnormality diagnosis system for mechanical equipment that can perform abnormality diagnosis with high accuracy without erroneous detection.

In order to achieve the above object, the machine equipment abnormality diagnosis system according to the present invention is characterized by the following (1), (2), (3), (4), (5) and (6).
(1) An abnormality diagnosis system for diagnosing an abnormality in a bearing or a bearing-related member in a mechanical facility by detecting sound or vibration generated from the mechanical facility and analyzing the detection signal,
An envelope processing unit for obtaining an envelope of the detection signal;
An FFT unit for converting the envelope obtained by the envelope processing unit into a frequency spectrum;
A peak detector for smoothing the frequency spectrum obtained by the FFT unit by moving average processing and detecting the peak;
A diagnosis unit for diagnosing an abnormality based on the peak of the frequency spectrum detected by the peak detection unit;
Having provided.
(2) In the abnormality diagnosis system configured as described in (1) above, the peak detection unit performs a smoothing differential process on the frequency spectrum obtained by the FFT unit, and the sign of the obtained differential value changes. A smoothing differential peak extraction unit that extracts frequency points to be extracted as frequency spectrum peaks.
(3) In the abnormality diagnosis system having the above configuration (1) or (2), the weighting coefficient in the moving averaging process is symmetrical (target before and after the current time).
(4) In the abnormality diagnosis system configured as described in (2) or (3) above, the peak detection unit has a root mean square of amplitude levels equal to or greater than a threshold value among the peaks extracted by the smoothed differential peak extraction unit A first sorting unit for sorting things is provided.
(5) In the abnormality diagnosis system configured as described in (4) above, the peak detection unit is configured to select a peak from a larger root mean square of amplitude levels to a predetermined number of peaks selected by the first selection unit. A second sorting unit for sorting the peaks is provided.
(6) In the abnormality diagnosis system according to any one of (1) to (5) above, the diagnosis unit includes a peak or vibration corresponding to a main component of vibration among the peaks of the frequency spectrum detected by the peak detection unit. The degree of coincidence between the peak corresponding to the main component and the higher-order component and the frequency indicating the abnormality of the diagnosis target is obtained, and the abnormality is diagnosed by evaluating the cumulative result of the coincidence multiple times.

According to the abnormality diagnosis system configured as described in (1) above, sound or vibration generated from mechanical equipment is detected, an envelope of the detection signal is obtained, the envelope is converted into a frequency spectrum, and the obtained frequency spectrum is moved. Since the peak is detected after smoothing by averaging, and abnormality is diagnosed based on the detected peak, even under conditions where the S / N ratio between the abnormal signal or abnormal sign signal and the noise signal is small The abnormality diagnosis can be performed with high accuracy without erroneously detecting the noise signal as an abnormality or an abnormal sign signal.
According to the abnormality diagnosis system having the configuration of (2) above, smoothed differential processing (that is, the sum of products of differences and lengths of a plurality of sections centered on the same point) is performed on the frequency spectrum, and the differential value is obtained. Since the frequency point where the sign of is changed is extracted as the peak of the frequency spectrum, the peak of the frequency spectrum buried in the noise can be detected with high accuracy.
According to the abnormality diagnosis system having the configuration (3) above, since the weighting coefficients in the moving averaging process are symmetrical, it is possible to prevent the noise signal from being erroneously detected as an abnormal signal or an abnormal sign signal.
According to the abnormality diagnosis system configured as described in (4) above, the extracted peaks whose root mean square of amplitude levels is greater than or equal to the threshold are selected, so that the peak detection of the frequency spectrum buried in the peak noise is further improved. It can be done with accuracy.
According to the abnormality diagnosis system having the configuration of (5) above, the peaks having the root mean square of the amplitude level larger than the threshold are selected from the peaks having the root mean square of the amplitude level greater than or equal to the threshold value. Abnormality diagnosis can be performed with high accuracy and efficiency by narrowing down to a peak effective for diagnosis.
According to the abnormality diagnosis system configured as described in (6) above, the peak corresponding to the main component of the vibration or the peak corresponding to the main component and higher order component of the vibration and the abnormality of the diagnosis target are detected. Since the degree of coincidence with the indicated frequency is obtained and the abnormality is diagnosed by evaluating the cumulative results of the coincidence for a plurality of times, the abnormality diagnosis can be performed with high accuracy.

  According to the present invention, even when the S / N ratio between the abnormal signal or the abnormal sign signal and the noise signal is small, the abnormality diagnosis is performed with high accuracy without erroneously detecting the noise signal as an abnormal or abnormal signal. it can.

  Hereinafter, the best mode for carrying out the present invention will be described by taking as an example the case of determining whether there is an abnormality such as a scratch on a rolling bearing in a mechanical facility, targeting a mechanical facility including a rolling bearing.

  FIG. 1 is a block diagram showing an example of an abnormality diagnosis system according to the present invention.

  As shown in FIG. 1, the abnormality diagnosis system of the present invention includes an amplifier / filter (filter processing unit) 1, an A / D converter 2, an envelope processing unit 3, an FFT unit 4, a peak detection unit 5, a diagnosis unit 6, And a diagnostic result output unit 7.

  The amplifier / filter 1 receives a signal detected by a sensor (vibration sensor, acoustic sensor, etc.) that detects sound or vibration generated from the machine equipment to be diagnosed. The amplifier / filter 1 amplifies an input signal with a predetermined gain and blocks a signal having a predetermined frequency or higher.

  The A / D converter 2 samples the analog signal that has passed through the amplifier / filter 1 at a predetermined sampling frequency and converts it into a digital signal.

  The envelope processing unit 3 obtains the envelope (envelope waveform) of the digital signal generated by the A / D converter 2.

  The FFT unit 4 analyzes the frequency of the envelope obtained by the envelope processing unit 3 and converts it into a frequency spectrum.

  The peak detector 5 detects the peak of the frequency spectrum obtained by the FFT unit 4.

  The diagnosis unit 6 compares the characteristic frequency determined by the rotational speed detected by a rotation sensor (not shown) provided in the rolling bearing and the internal specifications of the bearing with the peak obtained by the peak detection unit 5 and matches Diagnose the abnormality by evaluating the degree.

  The diagnosis result output unit 7 outputs the diagnosis result obtained by the diagnosis unit 6.

  The peak detection unit 5 includes a moving average processing unit 5a, a smoothed differential peak extraction unit 5b, a first sorting unit 5c, and a second sorting unit 5d.

一般には、次式（１）の演算を施すことにより、

周波数スペクトルを平滑化して雑音の軽減を行なう。 The moving averaging processing unit 5a weights the frequency spectrum (frequency domain discrete data) obtained by the FFT unit 4 symmetrically and performs moving averaging. For example, in a 5-point moving average, the following equation is applied to the frequency spectrum obtained by the FFT unit 4,

In general, by applying the following equation (1):

Noise is reduced by smoothing the frequency spectrum.

  The smoothed differential peak extraction unit 5b further smoothes the moving averaged spectrum after the moving averaging process by the moving average processing unit 5a to obtain a differential value, and obtains a frequency point at which the sign of the differential coefficient changes as a frequency spectrum. Extracted as a peak.

That is, the smoothed differential peak extraction unit 5b regards a frequency point at which the value of the following equation (2) (smoothed differential coefficient yj) changes from positive to negative as a candidate for a frequency spectrum peak.

  As can be seen from this equation (2), it can be seen that the slope between the distant points is greater in weight than the adjacent data. The peak detection unit 5 performs smoothing differentiation on the frequency spectrum obtained by the FFT unit 4 by performing a product sum of the differences of a plurality of sections and the lengths of the sections with the j point as the center, as can be seen from Equation (2). The smoothed differential peak extraction unit 5b that performs processing and extracts the frequency point at which the sign of the obtained differential value changes as a peak of the frequency spectrum is provided.

  Therefore, according to Equation (2), it is possible to detect a peak buried in noise without using Equation (1), but it may be used in combination with Equation (1).

  The first sorting unit 5c sorts out peaks extracted by the smoothed differential peak extracting unit 5b and whose root mean square of amplitude levels is greater than or equal to a threshold value. As the threshold value, a relative value determined according to the power average value or the root mean square of the peaks extracted by the smoothed differential peak extraction unit 5b is used. The absolute threshold is effective when the relative noise level is low, but is not necessarily effective when the noise level is high.

  The second sorting unit 5d sorts out the peaks sorted by the first sorting unit 5c up to a predetermined number from the larger root mean square of the amplitude level. As the simplest method, for example, a known sorting algorithm can be used to sort a plurality of peaks in descending or ascending order with respect to the level, and then sort them in descending order, that is, in descending order of value.

FIG. 2 shows an example of a frequency spectrum waveform. This example shows a spectrum diagnosed as having a flaw and a moving average processing result thereof. The moving average here is a moving average of 7 points as shown in the following equation.

  The weighting factor w is not limited to the above value, but it is desirable not to remove the condition that the weight of a point that is symmetric with respect to j = 0 and j = 0 is maximized.

  In the example of FIG. 2, since the S / N ratio is relatively good, the fundamental wave component f1 and the harmonic components f2, f3, and f4 due to scratches on the bearing outer ring are clearly seen before and after the moving average process. It can be seen that the number of false peaks due to noise is extremely small after the averaging process.

As shown in FIG. 2, the moving averaged spectrum is smoothed and differentiated by the moving average processing unit 5a, and the frequency point at which the sign of the differential coefficient changes from positive to negative is detected as a peak by the smoothed differential peak extracting unit 5b. Thereafter, the first selection unit 5c extracts those above the threshold, sorts them by the second selection unit 5d, and extracts the top five of them as peaks, whereby the peak frequencies f1, f2, f3, f4 are extracted. Is required. The smoothing differential coefficient yi at that time is expressed by the following equation, where xi is the discretized frequency spectrum.

  Unlike ordinary numerical differentiation, this formula gives a smoothing effect so that a greater weight is given to the difference between more distant points, so differentiation can be done only with integer arithmetic, and division is also possible. unnecessary. Therefore, even a microcomputer without a floating point arithmetic unit (FPU) or a division instruction can be operated without difficulty.

  The peak data of the frequency spectrum (envelope frequency distribution) obtained by the second selection unit 5d as described above is input to the diagnosis unit 6.

  The diagnosis unit 6 compares the peak corresponding to the main component of vibration or the peak corresponding to the main component and higher order component of the input frequency spectrum with the frequency indicating abnormality of the diagnosis target, and matches Find the degree. Then, a highly reliable diagnosis is performed by adding points to the obtained degree of coincidence and accumulating. For example, the main component, the second and fourth components are compared with the frequency indicating abnormality, and if the main component and other components are detected, it is determined that there is a possibility that a scratch has occurred. Then, the number of corresponding points in the preset score table is added. An example of the score table is shown in Table 1 below. In the example of FIG. 2, since the main component, the second order, and the third order component are detected, four points are added.

  In the example of the frequency spectrum waveform shown in FIG. 3, the frequency peak due to the scratch on the outer ring of the bearing is extracted while receiving noise due to external impact. As in the case of FIG. 2, smoothing differentiation is performed and peak detection is performed, and then the result is sorted as a peak by extracting values above the threshold to detect the main component and the secondary component. ing. In this case, the number of addition points is two points.

  In the example of the frequency spectrum waveform shown in FIG. 4, no noise is detected due to excessive noise due to external impact. In this case, the number of addition points is 0.

  FIG. 5 shows an example of a vibration waveform when impact noise is entered. The frequency analysis result of the envelope of the vibration waveform having such a large amplitude and sudden shocking noise becomes large on the low frequency side close to the DC (direct current) component, and is small as in the example of FIG. The peak of vibration due to scratches is hidden. In such a case, there is no need to forcibly perform processing for detecting a signal component due to a flaw.

  As shown in FIG. 6, this abnormality diagnosis system repeats a series of processes from the above-described vibration signal detection to abnormality point number determination a predetermined number of times N (for example, 30 times), and accumulates the number of points. The abnormality diagnosis is performed by. In FIG. 6, n is the current number, PA is a diagnostic point in one spectrum measurement, and PACC is a cumulative value of PA.

  The time required to perform abnormality diagnosis by sampling the frequency spectrum waveforms illustrated in FIGS. 2, 3, and 4 once is about 1 second. Therefore, if the time allowed for obtaining the diagnosis result is about 40 to 60 seconds, it is possible to repeat the diagnosis about 40 to 60 times, accumulate the above points, and perform an abnormality diagnosis based on the accumulated points. It is. In abnormality diagnosis by sampling only once, it is unclear what kind of spectrum is obtained as shown in FIGS. 2 to 4, but the frequency peak detection is repeated and the number of diagnosis points is added each time, By evaluating the cumulative value of the number of points, an abnormality diagnosis can be performed with high accuracy by reducing the influence of spectrum variation.

  FIG. 7 shows a diagnostic spectrum of a normal bearing without a flaw, and shows a measurement result in which a frequency component of vibration due to a flaw was not detected as a result of performing peak detection. The moving averaged frequency analysis result seems to have some features at first glance, but the result of selection processing by threshold and sorting is not related to the frequency component due to bearing abnormality, so the abnormality diagnosis points in Table 1 are not added. .

  FIG. 8 is a bar graph showing the cumulative number of diagnosis points after 40 times of abnormal diagnosis of a bearing with a minute flaw and a normal product. It can be seen that there is a large difference in the number of accumulated points between the micro-flawed product and the normal product, so that it is possible to accurately diagnose the bearing abnormality by accumulating the accumulated points for about 40 times. In addition, since a large difference is generated between the normal product and the fine product even though it is a minute scratch, the threshold range can be increased as shown in FIG. It is also possible to do so.

  As described above, in the abnormality diagnosis system of this embodiment, sound or vibration generated from mechanical equipment is detected, an envelope of the detection signal is obtained, the envelope is converted into a frequency spectrum, and the obtained frequency spectrum is converted into a frequency spectrum. After moving average processing, the spectrum is smoothed and differentiated, and the frequency point at which the sign of the derivative changes from positive to negative is detected as a peak, then those above the predetermined threshold are extracted and sorted. The peak number corresponding to the main component of the vibration or the peak corresponding to the main component of vibration and the higher-order component is matched with the frequency indicating the abnormality of the diagnosis target. The degree of coincidence is scored, accumulated several times, and the accumulated value is evaluated to diagnose the abnormality. Even under conditions S / N ratio is small and or abnormal sign signal and the noise signal, without detecting false noise signals and abnormality or abnormal sign signal can be performed abnormality diagnosis very and high efficiency with high accuracy.

  The present invention is not limited to the above embodiment. For example, as shown by a broken line block in FIG. 1, a digital filter (LPF / HPF) 8 is provided between the A / D converter 2 and the envelope processing unit 3 to remove a high-frequency noise component and reduce a DC offset. It is desirable to exclude. Further, a decimation unit 9 may be provided in front of the FFT unit 4 so as to perform a thinning process (decimation) according to a necessary frequency. By performing signal decimation processing after envelope processing to reduce the number of FFT calculation points for envelope waveform analysis, both improvement in frequency resolution of detected signals and improvement in FFT calculation efficiency are achieved. The bearing abnormality diagnosis can be performed with high accuracy and high efficiency.

It is a block diagram which shows the example of a form of the abnormality diagnosis system of this invention. It is a wave form diagram which illustrates a frequency spectrum and its moving average processing result. It is a wave form diagram which illustrates a frequency spectrum and its moving average processing result. It is a wave form diagram which illustrates a frequency spectrum and its moving average processing result. The example of the vibration waveform when impact noise enters is shown. It is a flowchart which shows the example of abnormality diagnosis operation | movement of the abnormality diagnosis system shown in FIG. It is a wave form diagram which illustrates a frequency spectrum and its moving average processing result. It is a figure which shows the abnormality diagnosis result of the micro flaw product of a bearing, and a normal product.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Amplifier filter 2 A / D converter 3 Envelope processing part 4 FFT part 5 Peak detection part 6 Diagnosis part 7 Diagnosis result output part

Claims (6)

An abnormality diagnosis system for diagnosing an abnormality in a bearing or a bearing-related member in a mechanical facility by detecting sound or vibration generated from the mechanical facility and analyzing the detection signal,
An envelope processing unit for obtaining an envelope of the detection signal;
An FFT unit for converting the envelope obtained by the envelope processing unit into a frequency spectrum;
A peak detection unit for smoothing the frequency spectrum obtained by the FFT unit by moving average processing and detecting the peak;
A diagnosis unit for diagnosing an abnormality based on the peak of the frequency spectrum detected by the peak detection unit;
An abnormality diagnosis system for mechanical equipment, characterized by comprising:   The peak detection unit performs a smoothing differential process on the frequency spectrum obtained by the FFT unit, and extracts a frequency point at which the sign of the obtained differential value changes as a peak of the frequency spectrum. The abnormality diagnosis system for mechanical equipment according to claim 1, further comprising an extraction unit.   The machine equipment abnormality diagnosis system according to claim 1, wherein weight coefficients in the moving averaging process are symmetrical.   The peak detection unit includes a first selection unit that selects, from among peaks extracted by the smoothed differential peak extraction unit, a peak whose root mean square of amplitude levels is equal to or greater than a threshold value. Item 4. The machine equipment abnormality diagnosis system according to Item 2 or 3.   The peak detection unit includes a second sorting unit that sorts up to a predetermined number of peaks having a larger root mean square of amplitude levels from among the peaks sorted by the first sorting unit. The machine equipment abnormality diagnosis system according to claim 4, wherein:   The diagnosis unit determines the degree of coincidence between the peak detected by the peak detection unit and the peak corresponding to the main component of vibration or the peak corresponding to the main component of vibration and the higher-order component and the frequency indicating abnormality of the diagnosis target. The abnormality diagnosis system for machine equipment according to any one of claims 1 to 5, wherein abnormality is diagnosed by obtaining and evaluating a cumulative result of a plurality of coincidences.

Images