TWI762221B - Detection method for rotation speed - Google Patents

Abstract

A detection method for rotation speed is provided. The detection method includes steps of: providing a time-domain vibration signal of a rotating device and converting the time-domain vibration signal into a frequency-domain vibration signal; processing the frequency-domain vibration signal by a signal processing method with a filter signal to obtain a weighted harmonic energy signal that is within a specific frequency range, wherein the weighted harmonic energy signal contains a plurality of wave peaks; and selecting a peak associated with highest weighted harmonic energy or probability from the plurality of wave peaks, wherein the frequency corresponding to the peak determines the rotation speed of the rotating device.

Description

Speed detection method

The present invention relates to the technical field of device detection, in particular to a rotational speed detection method of a rotating device.

The abnormal detection of rotating devices by equipment technicians is often supplemented by rotational speed information. Mainly, the frequency of the vibration characteristic of the rotating device is mostly related to the rotational speed. In addition to the originally installed vibration sensor, in order to be able to know the rotational speed of the rotating device, an additional rotational speed sensor is usually required. This not only occupies a signal acquisition channel, but also increases the cost of construction and maintenance.

Therefore, it is necessary to provide a rotational speed detection method to solve the problems existing in the conventional technology.

The main purpose of the present invention is to provide a rotational speed detection method, which processes and analyzes the vibration signal of the rotating device through a series of algorithms to obtain the rotational speed of the rotating device, which is beneficial to reduce the construction of the rotation sensor and reduce the abnormal detection of the rotating device. the cost of.

In order to achieve the above-mentioned purpose, the present invention provides a rotational speed detection method, comprising the following steps: S1: provide a time-domain vibration signal of a rotating device and convert the time-domain vibration signal into a frequency-domain vibration signal; S2: through a signal processing A method and a filter signal for processing the frequency domain vibration signal to obtain a weighted harmonic energy signal in a specific frequency range, wherein the weighted harmonic energy signal contains a plurality of peaks; and S3: from the plurality of peaks A highest peak is selected from among them, wherein the frequency corresponding to the highest peak determines the rotational speed of the rotating device.

In an embodiment of the present invention, the step S3 further includes: selecting the highest peak according to the peak feature vectors of the plurality of peaks and a peak feature model.

In an embodiment of the present invention, before the step S1, it also includes a step S0: implementing a modeling method to establish the wave peak characteristic model, the modeling method includes the following steps: The vibration signal is converted into a plurality of sample frequency domain vibration signals; each sample frequency domain vibration signal is processed through the signal processing method and the corresponding filter signal to obtain a sample weighted harmonic energy signal in a specific frequency range, wherein Each sample weighted harmonic energy signal contains multiple peaks; according to the characteristics of multiple peaks of each sample weighted harmonic energy signal, a peak feature vector is established; all the peak feature vectors and a rotational speed correlation vector are integrated to establish a peak feature model .

In one embodiment of the present invention, the signal processing method includes the following steps: determining a predetermined rotational speed range of the frequency domain vibration signal and determining the characteristics of the filter signal; implementing an oversampling method on the frequency domain vibration signal to obtain a high sampling signal; perform linear interpolation on the high sampling signal to obtain a peak probability signal; perform convolution calculation on the filter signal and the high sampling signal to obtain a harmonic energy signal; The filter signal and the peak probability signal are convolved to obtain a harmonic probability signal; the harmonic energy signal and the harmonic probability signal are convolved to obtain the weighted harmonic energy signal.

In one embodiment of the present invention, the step of establishing the peak characteristic vector further includes: characterizing multiple peaks of the weighted harmonic energy signal of each sample according to multiple peak characteristic formulas to obtain characterization data; A signal selection method is performed on the characterization data to obtain a peak eigenvalue; a peak eigenvector of each sample weighted harmonic energy signal is generated according to the peak eigenvalue.

In one embodiment of the present invention, the signal selection method includes supervised signal characteristic selection and unsupervised signal characteristic selection, wherein the supervised signal characteristic selection includes analysis of variance and mutual information.

In an embodiment of the present invention, the step of establishing the peak feature vector further includes normalizing the characterization data.

In an embodiment of the present invention, the step of determining a predetermined rotational speed range of the frequency domain vibration signal further comprises: determining the predetermined rotational speed range by user experience or using a cepstral method range of rotation.

In an embodiment of the present invention, the characteristics of the filter signal include interval, resolution and repetition times of the filter signal.

In an embodiment of the present invention, the oversampling method includes local spectral upsampling and spectral curve interpolation sampling, wherein the local spectral upsampling determines the sampled spectral range according to the filter signal.

Through the above detection method, the rotational speed of the device can be obtained by analyzing the vibration signal of the device under the condition that the rotational speed of the device does not change much, thereby improving the automation degree of fault diagnosis of the device and reducing the cost of abnormal detection of the device.

In order to make the above-mentioned and other objects, features and advantages of the present invention more clearly understood, the preferred embodiments of the present invention will be exemplified below and described in detail in conjunction with the accompanying drawings.

Please refer to Figure 1A. FIG. 1A is a flowchart of a rotational speed detection method according to an embodiment of the present invention. In one embodiment, the rotational speed detection method includes four steps S0-S3. Step S0 includes: implementing a modeling method to establish a peak characteristic model. Step S1 includes: providing a time-domain vibration signal of the rotating device and converting the time-domain vibration signal into a frequency-domain vibration signal. Step S2 includes: processing the frequency domain vibration signal through a signal processing method and a filter signal to obtain a weighted harmonic energy signal within a specific frequency range, wherein the weighted harmonic energy signal contains a plurality of peaks. Step S3 includes: selecting the highest wave crest from the plurality of wave crests, wherein the frequency corresponding to the highest wave crest determines the rotational speed of the rotating device.

First perform step S0: implement a modeling method to establish a wave crest feature model. Referring to the flowchart of Fig. 1B, the modeling method includes the following 4 steps: converting a plurality of sample time-domain vibration signals of known rotational speeds into a plurality of sample frequency-domain vibration signals; processing the signal through a signal processing method and a corresponding filter Each sample frequency domain vibration signal is used to obtain the sample weighted harmonic energy signal in a specific frequency range, wherein each sample weighted harmonic energy signal contains multiple peaks; according to each sample weighted harmonic energy signal multiple The characteristics of the wave crest are used to establish the wave crest characteristic vector; all the wave crest characteristic vectors and the speed correlation vector are integrated to establish the wave crest characteristic model.

The reason for implementing the modeling method is that the vibration frequency of the rotating device is related to the rotational speed. If a model can be established by combining the correlations between multiple vibration frequencies and the rotational speed, the rotational speed of the rotating device will be detected in the future. , as long as the vibration signal measured by the rotating device is imported into the model, the corresponding rotational speed can be obtained.

In the first step of the modeling method, a plurality of sample time-domain vibration signals are collected, each time-domain vibration signal corresponds to a known rotational speed, and each sample time-domain vibration signal is converted into a sample frequency domain by Fourier transform Vibration signal.

In the second step of the modeling method, each sample frequency domain vibration signal is processed through a signal processing method and a corresponding filter signal to obtain a sample weighted harmonic energy signal in a specific frequency range, respectively. Referring to Figure 8, Figure 8 is an example of a weighted harmonic energy signal. The signal processing method mentioned here is the same as the signal processing method mentioned later in step S2. In this embodiment of the present invention, the frequency domain signal that has undergone Fourier transform will obtain a weighted harmonic energy signal through a signal processing method. Here, these weighted harmonic energy signals (all corresponding to known rotational speeds) are the reference materials for forming the peak characteristic model in the modeling process. The rotational speed cannot be effectively and directly derived from the observation or analysis of the frequency domain signal, but the rotational speed can be effectively derived from the observation or analysis of the weighted harmonic energy signal (frequency domain signal after signal processing). Therefore, if a relationship can be established between the weighted harmonic energy signal and the rotational speed (referred to as the peak characteristic model here), in the future, as long as the weighted harmonic energy signal corresponding to the unknown rotational speed is imported into the peak characteristic model, it can be quickly obtained. The output of the speed.

With reference to the flow chart of Fig. 1C, the signal processing method comprises the following 6 steps: determine the preset rotational speed range of the frequency-domain vibration signal and determine the characteristics of the filter signal; The frequency-domain vibration signal is implemented by an oversampling method to obtain a high sampling signal ; Linearly interpolate the high sampling signal to obtain the peak probability signal; carry out the convolution calculation of the filter signal and the high sampling signal to obtain the harmonic energy signal; carry out the convolution calculation of the filter signal and the peak probability signal to obtain Obtain the harmonic probability signal; perform the convolution calculation of the harmonic energy signal and the harmonic probability signal to obtain the weighted harmonic energy signal.

First, the first step of signal processing is described: determining the preset rotational speed range of the frequency domain vibration signal and determining the characteristics of the filter signal. When modeling the process, the speed is known, so a preset speed range can be given. When the rotational speed is unknown, a preset rotational speed range needs to be given by user experience or a Cepstrum method is used to determine the predetermined rotational speed range. Since the rotating speed of the rotating device will present high energy in the form of equal intervals in a certain position of the frequency spectrum, the frequency domain vibration signal can show this phenomenon after passing through the cepstrum. For example, Figure 2 is an example of a derivative spectrum. In this example, the horizontal axis is time, the inverse of time is frequency, and the vertical axis is energy. A peak with a higher energy indicates that the corresponding frequency has a higher chance of appearing at equal intervals in the spectrum. In this example, the peaks with high energy are marked, and it can be observed that the peaks fall between 0.084 and 0.12 seconds, and the corresponding frequencies fall between 8.3 and 12 Hz (that is, the range of rotational speed).

In this step, the characteristics of the filter signal (called comb filter) are also determined according to a frequency range. The characteristics of the filter signal include the interval, resolution and repetition of the filter signal. For example, taking Figure 3 as an example, the frequency interval is 1.1Hz, the resolution is 0.1Hz (one sample every 0.1Hz, so 11 samples can be generated in one interval), and the number of repetitions is 5. Similarly, if the aforementioned frequency range of 8.3-12 Hz is used as an interval and the resolution is 0.07 Hz, the number of samples in an interval is 54, and the number of repetitions can be defined by the user.

In the second step of signal processing, the frequency domain vibration signal is subjected to oversampling to obtain an oversampled signal. Oversampling methods include local spectral upsampling (zoom FFT) and spectral curve interpolation (spline interpolation FFT). Referring to Figure 4, the signal at the top of the figure is the time domain vibration signal, the signal in the center of the figure is the converted frequency domain vibration signal and the high sampling signal after local frequency amplification and sampling, and the signal at the bottom of the figure is the converted frequency domain. Vibration signal and upsampled signal after spectral curve interpolation sampling. If the user wants to specify a specific spectral range, use partial spectral upsampling. The local spectrum upsampling is to determine the spectrum range after sampling according to the aforementioned filter signal. For example, a local spectral signal is obtained by convolving a signal in the frequency domain with the filter signal. For example, as shown in Figure 5, the upper part of the figure is a frequency domain vibration signal, the center of the figure is a filter signal, and the lower part of the figure is the result of the convolution calculation of the frequency domain vibration signal and the filter signal. It can be seen that the candidate frequency band of the signal is between 60 and 110 Hz.

In the third step of signal processing, the upsampled signal is linearly interpolated to obtain the peak probability signal. As shown in Figure 6, the upper part is a high sampling signal and the lower part is the peak probability signal. In this example, assuming that the candidate frequency band is 0~160Hz, the peak of the high sampling signal is detected, and the probability is set to 1 corresponding to the position of the peak, and the rest is defined by interpolation, such as linear interpolation. is 0 or 1.

In the fourth step of signal processing, the convolution of the filter signal and the upsampled signal is performed to obtain the harmonic energy signal. As shown at the top of Figure 7, the harmonic energy signal of Figure 7 can be obtained by convolving the filter signal with the high sampling signal at the top of Figure 6. In this example, the frequency range of the filter signal is 8Hz~12Hz, and the resolution is 0.01. In the fifth step of signal processing, the convolution of the filter signal and the peak probability signal is performed to obtain the harmonic probability signal. As shown at the bottom of Fig. 7, similarly, in this example, the frequency range of the filter signal of the filter signal is 8Hz~12Hz, and the resolution is 0.01. By convolving this filter signal with the peak probability signal at the bottom of Figure 6, the harmonic probability signal at the bottom of Figure 7 can be obtained.

In the sixth step of signal processing, the harmonic energy signal and the harmonic probability signal are convolved to obtain the weighted harmonic energy signal. In one embodiment, the weighted harmonic energy signal in FIG. 8 can be obtained by performing a convolution calculation on the harmonic energy signal at the top of FIG. 7 and the harmonic probability signal at the bottom of FIG. 7 . Comparing the signal in Figure 8 with the signal at the top of Figure 7, the peak with the highest peak is more prominent than the other peaks, which helps to correctly detect the highest peak in the weighted harmonic energy signal. Assuming that the weighted harmonic energy signal has only one peak, the frequency corresponding to this peak is the rotational speed. When the weighted harmonic energy signal has multiple peaks, but one peak is particularly prominent and the highest peak can be easily found, usually the frequency corresponding to the highest peak can determine the rotational speed. When the highest peak is not obvious, it is necessary to further analyze the characteristics of the relatively high peak to determine the corresponding frequency/speed. Similarly, when the weighted harmonic energy signal corresponds to an unknown rotational speed, it is still possible to quickly determine the rotational speed if the peak value can be directly compared to find the highest peak. However, when the highest peak is not obvious, it is necessary to use the peak characteristic model described below to determine the rotational speed. It should be noted that the highest peak mentioned here is not limited to the peak with the largest peak. The highest peak can be determined according to one of the 13 types of peak characteristics in Table 1, such as the area of the peak, the center of mass of the peak, or the peak characteristic value obtained by a combination thereof.

Next, the third step of the modeling method is performed, and a peak feature vector is established according to the features of a plurality of peaks of the sample weighted harmonic energy signal. In one embodiment, the step of establishing the peak feature vector further includes: characterizing multiple peaks of each sample weighted harmonic energy signal according to multiple peak characteristic formulas to obtain characterization data; normalizing the characterization data. Perform the signal selection method according to the characterization data to obtain the peak eigenvalues; generate the peak eigenvectors of each sample weighted harmonic energy signal according to the peak eigenvalues.

(1)

(2) 其中

代表第i筆資料的第J個特徵，N表示資料(信號)數量，

與

表示對應此特徵所算出的平均值(means)與標準差(standard deviation)，

表示正規化後的數據。 表1 時域 1.面積Area

2.質心Centroid

3.正規化質心

4.變異數Variance

5.不對稱Asymmetry

6.尾係數Tailing

7.二至四階矩Moment

8.偏度Skewness

9.峭度Kurtosis

10.對稱Symmetry

11.正規化峭度

12正規化指標

13.第K個峰與最高峰差值

For example, the weighted harmonic energy signal has multiple relatively high peaks. Taking Fig. 9 as an example, ƒ _i represents the frequency, and I(ƒ _i ) corresponds to the weighted harmonic energy of ƒ _i . Feature extraction is performed on each peak, for example, each peak is characterized according to the multiple peak characteristic formulas in Table 1 to obtain characterization data, and the characterization data is normalized. Normalization is done according to the following formulas (1)(2).

(1)

(2) of which

represents the J-th feature of the i-th data, N represents the number of data (signals),

and

Represents the mean and standard deviation calculated for this feature,

Represents normalized data. Table 1 Time Domain 1. Area

2. Centroid

3. Normalize the centroid

4. Variance

5. Asymmetric Asymmetry

6. Tailing coefficient

7. Moment of second to fourth order

8. Skewness

9. Kurtosis

10. Symmetry

11. Normalized kurtosis

12 Normalization Metrics

13. The difference between the Kth peak and the highest peak

Then, a signal selection method is performed according to the normalized data to obtain peak characteristic values. Signal selection methods include supervised signal feature selection and unsupervised signal feature selection.

個資料，每筆資料均具有以上描述子X。對於上述的任一描述子X，如果要了解其對分辨以上類別資料是否有顯著鑑別度，可以採F統計(F-statistic)計算其是否有顯著性。舉例來說，假設想知道以上描述子X中的某一個描述子對Y是否有足夠鑑別力，對某一資料

而言，其中i=1~K且j=1~

，假設N=

。對於第i群資料而言，平均值為

，對所有K群資料平均組內變異(with-group variablility)為

。令所有資料

總平均為

，則資料平均組間變異(between-group variability)為

。根據統計理論，對於f=B/W，其為具(K-1,N-K)自由度之F分佈之變數。為檢測描述子Y具有足夠鑑別力，可以先設定某一信心水準(confidence level)，例如95%，檢查f是否超過F(95%,K-1,N-K)這個門檻值，如果是的話，表示描述子Y具有足夠鑑別力區分這K群資料。若採用相互資訊法，各單一描述子X需先離散化成，例如L個區間，(L＞1)。用下列公式(3)計算，保留數值大的描述子X。

(3) In terms of supervised signal characteristic selection, each single descriptor X can be any index in Table 1, and the descriptor Y can be speed, time or signal type (for example, the peak corresponding to the speed, the peak corresponding to the non-speed). , or the device is normal, abnormal state one, abnormal state two, etc.). Suppose the user divides Y into K categories or K intervals. Supervised signal feature selection also includes analysis of variance (ANOVA, analysis of variance) and mutual information (MI, Mutual Information). If the analysis of variance method is used, it is assumed that each category in the K-category data has

Each data has the above descriptor X. For any of the above descriptors X, if you want to know whether it has a significant degree of discrimination for distinguishing the above categories of data, you can use F-statistic to calculate whether it is significant. For example, suppose you want to know whether one of the above descriptors X has sufficient discriminative power for Y, and for a certain data

, where i=1~K and j=1~

, assuming N =

. For group i data, the mean is

, the average within-group variance for all K-group data is

. all information

The overall average is

, the mean between-group variability of the data is

. According to statistical theory, for f=B/W, it is the variable of the F distribution with (K-1,NK) degrees of freedom. In order to detect that the descriptor Y has sufficient discrimination, you can first set a certain confidence level, such as 95%, and check whether f exceeds the threshold value of F(95%, K-1, NK). Descriptor Y is sufficiently discriminative to distinguish the K groups of data. If the mutual information method is adopted, each single descriptor X needs to be discretized into, for example, L intervals, (L>1). Calculate with the following formula (3), and retain the descriptor X with a large value.

(3)

(4) 接著計算任二單一描述子

與

，看其平均相關係數(Correlation coefficient)或相互資訊是否過大(例如0.6)，若是，則將其中之一捨去。假設描述子

與

均有n筆資料，相關係數計算方式如以下公式(5):

(5) 表2為算任二單一描述子

與

的特徵相關係數絕對值(或是相互資訊)矩陣範例，矩陣裡的元素[s ^i,j]代表任二單一描述子

與

特徵相關係數絕對值。假設共有5個描述子且相關係數最大為1，對角線部分代表信號自己對自已的相似度，故其值為1，其餘部分則是信號與信號彼此之間的相似度，故小於1。 表2   X ¹ X ² X ³ X ⁴ X ⁵ X ¹ 1.0 0.8 0.7 0.6 0.5 X ² 0.8 1.0 0.9 0.8 0.8 X ³ 0.7 0.9 1.0 0.8 0.8 X ⁴ 0.6 0.8 0.8 1.0 0.9 X ⁵ 0.5 0.8 0.8 0.9 1.0 如果要在這5個描述子裡找到最冗餘(Redundant)的資訊，可以根據下列公式(6)(7)用平均最大或是最小中最大方法將描述子移除。例如，若以平均方式計算，則對描述子1~5與其他描述子平均相似度分別為:2.6/5、3.3/5、3.2/5、3.1/5、3.0/5。因此描述子2(平均數值接近1)與其他描述子較相似，所以描述子2為優先捨去的描述子。當需要挑選多個描述子捨去時，可以先把平均相似度由大到小排序，然後挑選前幾大數值所對應的描述子進行捨去。

(6)

(7) For unsupervised signal feature selection, the following formula (4) is used to calculate whether the value entropy value of each single descriptor X is too small, and if it is too small (for example, less than 0.5), it is discarded.

(4) Then calculate any two single descriptors

and

, to see if the average correlation coefficient (Correlation coefficient) or mutual information is too large (eg 0.6), if so, discard one of them. hypothetical descriptor

and

There are n pieces of data, and the correlation coefficient is calculated as the following formula (5):

(5) Table 2 is the calculation of any two single descriptors

and

An example of the absolute value (or mutual information) matrix of the characteristic correlation coefficient of , where the elements [s ^i,j ] in the matrix represent any two single descriptors

and

The absolute value of the feature correlation coefficient. Assuming that there are 5 descriptors and the maximum correlation coefficient is 1, the diagonal part represents the similarity of the signal to itself, so its value is 1, and the rest is the similarity between the signal and the signal, so it is less than 1. Table 2 X ¹ X ² X ³ X ⁴ X ⁵ X ¹ 1.0 0.8 0.7 0.6 0.5 X ² 0.8 1.0 0.9 0.8 0.8 X ³ 0.7 0.9 1.0 0.8 0.8 X ⁴ 0.6 0.8 0.8 1.0 0.9 X ⁵ 0.5 0.8 0.8 0.9 1.0 If you want to find the most redundant (Redundant) information in these 5 descriptors, you can remove the descriptors by using the average maximum or the minimum medium maximum method according to the following formulas (6) (7). For example, if calculated in an average way, the average similarity between descriptors 1 to 5 and other descriptors is: 2.6/5, 3.3/5, 3.2/5, 3.1/5, 3.0/5. Therefore, descriptor 2 (with an average value close to 1) is more similar to other descriptors, so descriptor 2 is the one to be discarded preferentially. When multiple descriptors need to be selected and discarded, the average similarity can be sorted from large to small first, and then the descriptors corresponding to the first few large values can be selected and discarded.

(6)

(7)

:

Next, a peak eigenvector for each sample weighted harmonic energy signal is generated from the peak eigenvalues. After the above signal characteristic selection, it is assumed that n ₁ peak characteristic values of the wave peaks corresponding to the rotational speed and n ₀ wave crest characteristic values of the wave peaks not corresponding to the rotational speed are obtained, n ₁ + n ₀ =N. The peak eigenvalues of multiple peaks of each signal are established as a peak eigenvector as follows

:

可以描述如下:

當有了向量

與

之後，下一步便是找出二者的對應函式，利用簡單線性回歸(linear regression)來建立波峰特徵模型:

Finally, the third step of the modeling method is carried out, and all wave crest characteristic vectors and rotational speed correlation vectors are integrated to establish a wave crest characteristic model. It should be noted that after the eigenvalues of the wave peaks are obtained, they will be arranged in order according to the eigenvalues, for example, the numbers are 1, 2… in sequence. In addition, a speed-dependent vector can also be established. When the peak characteristic value can determine the speed, the corresponding vector value is marked as 1. Otherwise, when the peak characteristic value cannot determine the rotational speed, the corresponding vector value is marked as 0. The established speed correlation vector

It can be described as follows:

when there is a vector

and

After that, the next step is to find the corresponding function of the two, and use a simple linear regression to build a peak feature model:

When there is a peak of the weighted harmonic energy signal of unknown speed to be judged in the future, b=a*x can be calculated, where b=[0,1], which are respectively mapped to the scores of the corresponding speed and no corresponding speed, and the maximum score is taken. The wave peak corresponding to the value is used for speed output. In addition to simple linear regression analysis, other methods, such as neural network or support vector machine, can also be used to find out the corresponding relationship and store it as a model for subsequent prediction.

After the wave peak characteristic model is established, then step S1 of the present invention is performed: providing a time domain vibration signal of the rotating device and converting the time domain vibration signal into a frequency domain vibration signal, and step S2: using a signal processing method and a filter signal to generate The frequency domain vibration signal is processed to obtain a weighted harmonic energy signal in a specific frequency range, wherein the weighted harmonic energy signal contains a plurality of peaks. The above-mentioned time-domain vibration signal is measured from a rotating device with an unknown rotational speed (the rotational speed to be measured). The methods of the above steps S1 and S2 are similar to the methods of the first step and the second step of the modeling method. The related signal processing method is also similar to the signal processing method mentioned in the modeling method, and will not be repeated here.

After the weighted harmonic energy signal is obtained, step S3 is performed: the highest wave crest is selected from the plurality of wave crests, wherein the frequency corresponding to the highest wave crest determines the rotational speed of the rotating device. In one embodiment, if the weighted harmonic energy signal has only a single peak, the frequency corresponding to the single peak determines the rotational speed of the rotating device. In another embodiment, the highest peak can be found according to the peak size of these peaks, for example, the highest peak can be selected according to the largest peak. For example, as shown in Figure 8, one of the peaks in the weighted harmonic energy signal is particularly prominent, with a peak as high as 7.7, and most of the other peaks have peaks below 4. If the peak value is used for screening, the frequency corresponding to the peak of the maximum peak value of 7.7 is 10.48Hz, that is, the rotation speed is 10.48Hz.

In other embodiments, when the peaks of many peaks are close or the peak shapes are similar, the highest peak can be selected according to the peak feature vectors and peak feature models of the multiple peaks. It should be noted that the peak eigenvector of the weighted harmonic energy signal of the unknown rotational speed can be obtained by referring to the peak of the sample weighted harmonic energy signal in the modeling method by mapping. That is to say, if the peak characteristics of the weighted harmonic energy signal of the unknown rotational speed are similar to the characteristics of the peaks of one of the sample weighted harmonic energy signals, the weighted harmonic energy signal of the unknown rotational speed can be obtained by referring to that sample weighted harmonic energy signal. The peak eigenvectors of the wave energy signal. Finally, as long as the wave crest characteristic vector is imported into the wave crest characteristic model established above, the output result determines the rotational speed to be detected.

Through the above detection method, the rotational speed of the device can be obtained by analyzing the vibration signal of the device under the condition that the rotational speed of the equipment does not change much, thereby improving the automation degree of fault diagnosis of the rotating device and reducing the cost of abnormal detection of the rotating device.

The above examples are merely examples of the above principles. It should be understood that modifications and variations of the arrangements and details described herein will be apparent. Therefore, it is the intent of the present invention to be limited in scope by the present claims and not by the specific details presented by way of description and explanation of the examples herein.

S0-S3: Process steps

FIG. 1A is a flowchart of a rotational speed detection method according to an embodiment of the present invention. FIG. 1B is a flowchart of the modeling method of the present invention. FIG. 1C is a flowchart of the signal processing method of the present invention. Fig. 2 is a cepstrum of an embodiment of the present invention. FIG. 3 is a schematic diagram of a filter signal according to an embodiment of the present invention. FIG. 4 is a schematic diagram of a time domain signal and a frequency domain signal according to an embodiment of the present invention. FIG. 5 is an example of convolution calculation according to an embodiment of the present invention. FIG. 6 is a schematic diagram of a high sampling signal and a peak probability signal according to an embodiment of the present invention. FIG. 7 is a schematic diagram of a harmonic energy signal and a harmonic probability signal according to an embodiment of the present invention. FIG. 8 is a schematic diagram of a weighted harmonic energy signal according to an embodiment of the present invention. FIG. 9 is a schematic diagram of a local peak of a weighted harmonic energy signal according to an embodiment of the present invention.

S0-S3: Process steps

Claims (9)

A rotational speed detection method, comprising the following steps: S1: provide a time-domain vibration signal of a rotating device and convert the time-domain vibration signal into a frequency-domain vibration signal; S2: process the signal by a signal processing method and a filter signal frequency domain vibration signal to obtain a weighted harmonic energy signal within a specific frequency range, wherein the weighted harmonic energy signal contains a plurality of peaks, wherein the signal processing method includes: determining a preset rotational speed range of the frequency domain vibration signal and determine the characteristics of the filter signal; implement the oversampling method on the frequency domain vibration signal to obtain a high sampling signal; perform linear interpolation on the high sampling signal to obtain a peak probability signal; convolve the filter signal and the high sampling signal Calculate to obtain the harmonic energy signal; convolve the filter signal and the peak probability signal to obtain the harmonic probability signal; and perform the convolved calculation of the harmonic energy signal and the harmonic probability signal to obtain the weighted harmonic an energy signal; and S3 : selecting a highest peak from the plurality of peaks, wherein the frequency corresponding to the highest peak determines the rotational speed of the rotating device. The rotational speed detection method as claimed in claim 1, wherein the step S3 further comprises: selecting the highest peak according to peak feature vectors of the plurality of peaks and a peak feature model. The rotational speed detection method according to claim 2, further comprising a step S0 before the step S1: implementing a modeling method to establish the wave peak characteristic model, and the modeling method includes the following steps: Convert a plurality of sample time-domain vibration signals of known rotational speed into a plurality of sample frequency-domain vibration signals; process each sample frequency-domain vibration signal through the signal processing method and the corresponding filter signal to obtain a specific frequency range respectively. The sample weighted harmonic energy signal in the sample weighted harmonic energy signal, wherein each sample weighted harmonic energy signal contains multiple peaks; according to the characteristics of the multiple peaks of each sample weighted harmonic energy signal, a peak eigenvector is established; all the peak eigenvectors and A rotational speed correlation vector is used to establish the peak characteristic model. The rotational speed detection method according to claim 3, wherein the step of establishing a peak characteristic vector further comprises: characterizing a plurality of peaks of the weighted harmonic energy signal of each sample according to a plurality of wave peak characteristic formulas to obtain a characteristic data; perform a signal selection method according to the characterization data to obtain a peak feature value; generate a peak feature vector of each sample weighted harmonic energy signal according to the peak feature value. The rotational speed detection method as claimed in claim 4, wherein the signal selection method includes supervised signal characteristic selection and unsupervised signal characteristic selection, wherein the supervised signal characteristic selection includes variance analysis method and mutual information method. The rotational speed detection method as claimed in claim 4, wherein the step of establishing a peak feature vector further comprises normalizing the feature data. The rotational speed detection method as claimed in claim 1, wherein the step of determining a predetermined rotational speed range of the frequency domain vibration signal further comprises: The predetermined rotational speed range is given by user experience or determined by using a cepstral method. The rotational speed detection method as claimed in claim 1, wherein the characteristics of the filter signal include interval, resolution and repetition times of the filter signal. The rotational speed detection method as claimed in claim 1, wherein the oversampling method comprises partial spectrum upsampling and spectral curve interpolation sampling, wherein the partial spectrum upsampling determines the spectrum range after sampling according to the filter signal.

Images