Vibration prediction method based on expert knowledge and artificial intelligence

Abstract

The invention discloses a vibration prediction method based on expert knowledge and artificial intelligence. The method comprises the following steps: performing time slicing on a vibration time sequence, and calculating a factor of each time slice; obtaining a corresponding factor time sequence according to the factor of each time slice; establishing a prediction model, and predicting the vibration time sequence and the factor time sequence; calculating a residual sequence according to the predicted value and the actual value, and adjusting the prediction model according to the residual sequence to obtain a final prediction model; wherein the factors comprise a peak factor, a pulse factor, a margin factor and a kurtosis factor. An intelligent method is introduced, time sequence analysis processing is carried out on time domain and frequency domain signals, and the manual processing burden of personnel is relieved.

Description

Technical Field

[0001] The present invention relates to the field of vibration analysis, and more particularly, to a vibration prediction method based on expert knowledge and artificial intelligence. Background Art

[0002] Currently, the vibration analysis method based on expert knowledge judges the impact and wear conditions by calculating the relevant dimensionless time series signals of a vibration time series. It depends on expert experience. Although it realizes informatization, the observation operation needs to be manually processed and is difficult to implement.

[0003] Therefore, it is necessary to develop a vibration prediction method based on expert knowledge and artificial intelligence.

[0004] The information disclosed in the background art section of the present invention is only intended to deepen the understanding of the general background art of the present invention, and should not be regarded as an admission or any form of suggestion that this information constitutes the prior art known to those skilled in the art. Summary of the Invention

[0005] The present invention proposes a vibration prediction method based on expert knowledge and artificial intelligence, which can introduce intelligent methods to perform time series analysis and processing on time domain and frequency domain signals, and reduce the manual processing burden of personnel.

[0006] Embodiments of the present disclosure provide a vibration prediction method based on expert knowledge and artificial intelligence, including:

[0007] Performing time slicing on the vibration time series and calculating the factors of each time slice;

[0008] Obtaining the corresponding factor time series according to the factors of each time slice;

[0009] Establishing a prediction model to perform prediction on the vibration time series and the factor time series;

[0010] Calculating the residual sequence according to the predicted value and the actual value, and adjusting the prediction model according to the residual sequence to obtain the final prediction model;

[0011] Wherein, the factors include peak factor, impulse factor, margin factor, and kurtosis factor.

[0012] Preferably, before establishing the prediction model, it further includes:

[0013] Testing the stationarity of the factor time series. If it is non-stationary, the sequence is made stationary by the differencing method. If it is stationary, a prediction model is established.

[0014] Preferably, testing the stationarity of the factor time series includes:

[0015] Perform a stationarity test using ADF, calculate the test statistic. If the test statistic is less than the critical value, reject the null hypothesis H 0 , and accept the alternative hypothesis H 1 , that is, consider the time series to be stationary;

[0016] Among them, the null hypothesis H 0 is that the time series has a unit root, and the alternative hypothesis H 1 is that the time series does not have a unit root.

[0017] Preferably, adjusting the prediction model according to the residual sequence to obtain the final prediction model includes:

[0018] If the vibration time series, peak factor series, impulse factor series, margin factor series, and kurtosis factor series are all white noise, the current model is the final prediction model.

[0019] Preferably, adjusting the prediction model according to the residual sequence to obtain the final prediction model includes:

[0020] If at least one of the vibration time series, peak factor series, impulse factor series, margin factor series, and kurtosis factor series is not white noise, modify the prediction model.

[0021] Preferably, modifying the prediction model includes at least one of: performing a fitting attempt through automatic parameter tuning, changing the model, and making a secondary prediction of the residuals.

[0022] Preferably, the peak factor is:

[0023]

[0024] The effective value is:

[0025]

[0026] Preferably, the impulse factor is:

[0027]

[0028] The rectified average value is:

[0029]

[0030] Preferably, the margin factor is:

[0031]

[0032] The root mean square amplitude is:

[0033]

[0034] Preferably, the kurtosis factor is as follows:

[0035]

[0036] wherein

[0037] The method of the present invention has other characteristics and advantages, which will be apparent from the accompanying drawings incorporated herein and the subsequent detailed description, or will be described in detail in the accompanying drawings incorporated herein and the subsequent detailed description, and these accompanying drawings and detailed description are used together to explain the specific principles of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] By describing the exemplary embodiments of the present invention in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present invention will become more apparent, wherein, in the exemplary embodiments of the present invention, the same reference numerals generally represent the same components.

[0039] Figure 1 A flowchart showing the steps of a vibration prediction method based on expert knowledge and artificial intelligence according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0040] The preferred embodiments of the present invention will be described in more detail below. Although the preferred embodiments of the present invention are described below, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein.

[0041] To facilitate understanding of the solution and its effects of the embodiments of the present invention, a specific application example is given below. Those skilled in the art should understand that this example is only for facilitating the understanding of the present invention, and any specific details are not intended to limit the present invention in any way.

[0042] Example 1

[0043] Figure 1 A flowchart showing the steps of a vibration prediction method based on expert knowledge and artificial intelligence according to an embodiment of the present invention is shown.

[0044] As Figure 1 shown, the vibration prediction method based on expert knowledge and artificial intelligence includes:

[0045] Step 101: Perform time slicing on the vibration time series and calculate the factors of each time slice;

[0046] Step 102: Obtain the corresponding factor time series according to the factors of each time slice;

[0047] Step 103: Establish a prediction model to perform predictions on the vibration time series and the factor time series;

[0048] Step 104: Calculate the residual sequence based on the predicted values and the actual values, and adjust the prediction model according to the residual sequence to obtain the final prediction model;

[0049] Among them, the factors include the peak factor, the pulse factor, the margin factor, and the kurtosis factor.

[0050] In one example, before establishing the prediction model, it further includes:

[0051] Check the stationarity of the factor time series. If it is non-stationary, the sequence is made stationary by differencing. If it is stationary, establish the prediction model.

[0052] In one example, checking the stationarity of the factor time series includes:

[0053] Use ADF for stationarity testing and calculate the test statistic. If the test statistic is less than the critical value, reject the null hypothesis H 0 , and accept the alternative hypothesis H 1 , that is, consider the time series to be stationary;

[0054] Among them, the null hypothesis H 0 is that the time series has a unit root, and the alternative hypothesis H 1 is that the time series does not have a unit root.

[0055] In one example, adjusting the prediction model according to the residual sequence to obtain the final prediction model includes:

[0056] If the vibration time series, the peak factor sequence, the pulse factor sequence, the margin factor sequence, and the kurtosis factor sequence are all white noise, the current model is the final prediction model.

[0057] In one example, adjusting the prediction model according to the residual sequence to obtain the final prediction model includes:

[0058] If at least one of the vibration time series, the peak factor sequence, the pulse factor sequence, the margin factor sequence, and the kurtosis factor sequence is not white noise, modify the prediction model.

[0059] In one example, modifying the prediction model includes at least one of: performing fitting attempts through automatic parameter tuning, changing the model, and making a secondary prediction of the residuals.

[0060] In one example, the peak factor is:

[0061]

[0062] The effective value is:

[0063]

[0064] In one example, the pulse factor is:

[0065]

[0066] The rectified average value is:

[0067]

[0068] In one example, the margin factor is:

[0069]

[0070] The root mean square value is:

[0071]

[0072] In one example, the kurtosis factor is:

[0073]

[0074] Wherein,

[0075] Specifically, the vibration time series is first sliced in time. Considering that the vibration signal has a high frequency, the length n of the time slice here (due to the law of large numbers, it is defaulted that n is greater than or equal to 50, or it can also be set manually). That is, every time n data are collected, the corresponding peak factor, pulse factor, margin factor, and kurtosis factor of this vibration time series are calculated.

[0076] Suppose the vibration time series {X 1 , X 2 …, X N} is divided into m groups, with n in each group, where N = nm.

[0077] In this way, the sequences of the peak factor, pulse factor, margin factor, and kurtosis factor of the vibration time series changing with time can be obtained. Here, it is assumed that {X 1 , X 2 …, X N} is a time series, and this time series can be divided into n = N / m subsequences according to a certain length m. The length m of the time subsequence should be at least twice the prediction length. Then an n×m two-dimensional matrix is obtained:

[0078]

[0079] Where {X 1 , X 2 …, X NCorrespondence with (Matrix 1):

[0080] The first row of the matrix {Y 11 , Y 12 …, Y 1m} corresponds to {X 1 , X 2 …, X m}

[0081] The second row of the matrix {Y 21 , Y 22 …, Y 2m} corresponds to {X m+1 , X m+2 …, X 2m}

[0082] ……

[0083] The nth row of the matrix {Y n1 , Y n2 …, T nm} corresponds to {X N-m+1 , X N-m+2 …, X N}

[0084] Corresponding formula: Y ij = X (i-1)*m+j

[0085] The following correlation factors are for the subsequence, i.e., factor calculation for each row of the two-dimensional matrix:

[0086] The peak of the vibration signal is the maximum amplitude reached by the vibration signal within each row of (Matrix 1). The specific mathematical expression for this peak is where i represents the ith row of (Matrix 1). In this way, a peak sequence of the vibration signal can be obtained n peaks.

[0087] The crest factor is the ratio of the peak sequence of the vibration signal to the effective value sequence, representing the extreme degree of the vibration peak in the vibration waveform:

[0088]

[0089] The effective value is obtained by summing the squares of the subsequence of the ith row of the two-dimensional matrix, finding the mean, and then taking the square root:

[0090]

[0091] By calculating the effective value of the ith row of (Matrix 1), an effective value sequence can be obtained So the crest factor sequence

[0092] The pulse factor is the ratio of the peak of the vibration signal to the rectified average value:

[0093]

[0094] The rectified average value is the average value of the absolute values, which is to take the absolute values of the subsequences in the i-th row of the two-dimensional matrix, sum them up, and then average:

[0095]

[0096] By calculating the rectified average value of the i-th row of (Matrix 1), a rectified average value sequence can be obtained Therefore, the pulse factor sequence

[0097] The margin factor is the ratio of the peak of the vibration signal to the root mean square amplitude.

[0098] The root mean square amplitude is:

[0099]

[0100] By calculating the root mean square amplitude of the i-th row of (Matrix 1), a root mean square amplitude sequence can be obtained Therefore, the margin factor sequence

[0101] The kurtosis factor represents the flatness of the vibration signal waveform and is used to describe the distribution of the vibration signal. The formula for the kurtosis factor is:

[0102]

[0103] Among them,

[0104] Calculate the kurtosis factor of the i-th row of (Matrix 1) The kurtosis of the normal distribution is 3. A distribution with a value less than 3 is flat and has a small impact, while a distribution with a value greater than 3 is steep and has a large impact.

[0105] In this way, a peak factor time series {F 1 ,F 2 …,F m}, a pulse factor time series {M 1 ,M 2 …,M m}, a margin factor {Y 1 ,Y 2 …,Y m}, and a kurtosis factor time series {Q 1 ,Q 2 …,Q m} can be obtained.

[0106] Time series analysis can be performed on the peak factor, pulse factor, margin factor, and kurtosis factor. The specific steps are as follows:

[0107] (1) Check whether the time series (factor time series) is stationary. When performing the ADF test, a judgment is made by calculating the test statistic and comparing it with a specific critical value. If the calculated test statistic is less than the critical value, the null hypothesis H 0 is rejected, and the alternative hypothesis H 1 is accepted, that is, the time series is considered stationary. Here, the null hypothesis H 0 of the ADF test is: The time series has a unit root, that is, the time series is non-stationary. The corresponding alternative hypothesis H 1 is: The time series does not have a unit root.

[0108] (2) If the factor time series is non-stationary, use the differencing method to make the sequence stationary. First, perform the first difference, that is, the first-order difference Check again whether the factor time series is stationary. If it is still not stationary, continue to difference. Difference the first-order difference, that is, the second-order difference until the time series is stationary (usually at most two differences are done). If it is stationary, perform a white noise test on the vibration time series. If the sequence is white noise, no prediction model is needed.

[0109] (3) Predict the vibration time series and its factor time series. After prediction, calculate the residuals with the actual values. If the residual sequence is white noise, the model fits well, indicating that the regular and predictable data have been fitted into the model, and the remaining data is unpredictable. If the residuals are not white noise, it means the model needs to be further optimized. You can try to fit by automatic parameter adjustment. If it still doesn't work, try a different model, or perform a second prediction on the residuals. If all five residual sequences: the vibration time series, peak factor sequence, pulse factor sequence, margin factor sequence, and kurtosis factor sequence are white noise, it means the prediction model is completed.

[0110] Use the Ljung Box test (for small samples) for the white noise test. If the p-value is greater than 0.05, this time series is considered white noise. If the p-value is less than 0.05, the time series is non-white noise.

[0111] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.

[0112] The embodiments of the present invention have been described above. The above description is exemplary and not exhaustive, and is also not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A vibration prediction method based on expert knowledge and artificial intelligence, characterized in that: include: Time slice the vibration time series and calculate the factor of each time slice; According to the factors of each time slice, a corresponding factor time series is obtained; Establishing a prediction model to predict the vibration time series and the factor time series; Calculate a residual sequence according to the predicted value and the actual value, adjust the prediction model according to the residual sequence, and obtain a final prediction model; The factors include peak factor, pulse factor, margin factor and kurtosis factor.

2. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: Before building a prediction model, it also includes: The stationarity of the factor time series is tested. If it is non-stationary, the series is stabilized by difference method. If it is stationary, a prediction model is established.

3. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 2, wherein: Testing the stationarity of the factor time series includes: ADF is used to perform a stationarity test and calculate the test statistic. If the test statistic is less than the critical value, the null hypothesis H0 is rejected and the alternative hypothesis H1 is accepted, which means that the time series is considered to be stationary. Among them, the null hypothesis H0 is that the time series has a unit root, and the alternative hypothesis H1 is that the time series does not have a unit root.

4. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: Adjusting the prediction model according to the residual sequence to obtain the final prediction model includes: If the vibration time series, peak factor series, impulse factor series, margin factor series, and kurtosis factor series are all white noises, the current model is the final prediction model.

5. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: Adjusting the prediction model according to the residual sequence to obtain the final prediction model includes: If at least one of the vibration time series, the peak factor series, the impulse factor series, the margin factor series, and the kurtosis factor series is not white noise, the prediction model is modified.

6. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 5, wherein: Modifying the prediction model includes at least one of: performing a fitting attempt through automatic parameter adjustment, replacing the model, and performing a secondary prediction on the residual.

7. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: The peak factor is: Valid values ​​are:

8. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: The pulse factor is: The rectified mean is:

9. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: The margin factor is: The square root amplitude is:

10. The vibration prediction method based on expert knowledge and artificial intelligence according to claim 1, wherein: The kurtosis factor is: in,

Images