Method for dividing health state of aero-generator bearing

Abstract

The invention relates to an aero-generator bearing health state division method, which comprises the following steps: S0, acquiring an aero-generator bearing vibration signal and preprocessing the aero-generator bearing vibration signal to obtain health index data; the method comprises the following steps: S1, determining a data escape point as a first escape point according to distribution characteristics of health index data in a set division window; s2, the division window is expanded backwards by a set length, and a data escape point in the expanded division window is obtained to serve as a second escape point; and S3, judging whether the first escape point is a state demarcation point or not according to a comparison relationship between the first escape point and the second escape point, if not, returning to the step S2, otherwise, resetting the division window and returning to the step S1 until all data points of the health index data are traversed to obtain a state division result. According to the method, the health index data of each stage of the bearing life cycle can be accurately divided.

Description

Technical Field

[0001] The invention relates to the technical field of aircraft generator bearing detection, and in particular to a method for dividing the health status of an aircraft generator bearing. Background Art

[0002] According to the health management requirements of aircraft generators, the wear and degradation data of generator bearings are monitored. By analyzing the bearing degradation data, the health status of the bearings presents multi-stage characteristics. In order to further study the health assessment and life prediction of generators, it is necessary to pre-process the bearing degradation data, divide the health status of the bearings, separate the different stages of bearing degradation, and then build a health management model based on the degradation characteristics of each stage. At present, for the health management of aircraft bearings, the health status of the bearings is generally divided into the health stage and the degradation stage, and the degradation stage is not further divided into multi-stage degradation, namely the initial degradation stage, the steady degradation stage, and the accelerated degradation stage. In order to track the degradation stage of the bearing more accurately and improve the accuracy of the bearing health management model, it is necessary to further subdivide the degradation stage of the bearing to improve the level of refinement of the health management of aircraft generators. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for dividing the health status of an aircraft generator bearing, which can accurately divide the health indicator data of each stage of the bearing life cycle.

[0004] The technical solution adopted by the present invention to solve the technical problem is: to provide a method for dividing the health status of an aircraft generator bearing, comprising the following steps:

[0005] S0 obtains the vibration signal of the aircraft generator bearing and pre-processes it to obtain health index data;

[0006] S1 determines the data escape point as the first escape point according to the distribution characteristics of the health indicator data in the set partition window;

[0007] S2 extends the partition window backward by a set length, and obtains the data escape point in the expanded partition window as the second escape point;

[0008] S3 determines whether the first escape point is a state dividing point based on the comparison relationship between the first escape point and the second escape point. If it is not a state dividing point, it returns to step S2. Otherwise, it resets the division window and returns to step S1 until all data points of the health indicator data are traversed to obtain the state division result.

[0009] Furthermore, the step of determining the data escape point according to the distribution characteristics of the health indicator data in the set partition window includes:

[0010] Calculate the probability density of each data point in the divided window, and extract the data point with the largest probability density as the health indicator data point;

[0011] The first escape point is determined according to the distribution relationship of each data point in the divided window relative to the health indicator data point.

[0012] Furthermore, the probability density of each data point in the partitioned window is calculated by a kernel density estimator.

[0013] Furthermore, the determining of the first escape point according to the distribution relationship of each data point in the divided window relative to the health indicator data point includes:

[0014] Compare the current data point and the subsequent consecutive data points in the partition window with the health index data point. If they are all larger than the health index data point, the current data point is used as the first escape point.

[0015] Furthermore, it also includes:

[0016] If there is no first escape point in the current partition window, the partition window is extended backward by a set length and the steps of comparing the current data point and several consecutive data points thereafter in the partition window with the size of the health indicator data points are repeated until the first escape point is determined or all data points of the health indicator data are traversed.

[0017] Furthermore, the current state dividing point is used as the starting point when resetting the divided window.

[0018] Furthermore, the health indicator data is obtained by the following method:

[0019] A data window of a set length is used to slide along the vibration signal of the aircraft generator bearing with a set step length;

[0020] The effective value of the vibration signal of the aircraft generator bearing in each sliding window is calculated to obtain the health index data.

[0021] Furthermore, the method also includes the step of taking different processing methods for data at different stages according to the state division results.

[0022] Furthermore, different processing methods are adopted for data at different stages, including:

[0023] Data from stages with stationary characteristics are not used to build prediction models;

[0024] The data of the stages with trend characteristics are used to build a degradation prediction model.

[0025] Beneficial Effects

[0026] Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: by analyzing the probability density of the preprocessed aircraft generator bearing vibration signal, the present invention finds the escape point and determines whether the escape point is a state dividing point, and can accurately divide the various stages of the bearing life cycle health indicator data, extract the most suitable data for degradation prediction, and thus improve the pertinence and accuracy of subsequent degradation prediction models and life prediction models. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a logic diagram of dividing health status cutoff points according to an embodiment of the present invention;

[0028] Figure 2 is a health index construction diagram of an embodiment of the present invention;

[0029] Figure 3 is a health indicator data kernel density estimation diagram according to an embodiment of the present invention;

[0030] Figure 4 4 is a multi-stage division diagram of the bearing health status according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0032] The embodiment of the present invention relates to a method for classifying the health status of an aircraft generator bearing, such as Figure 1 As shown, the following steps are included:

[0033] S0 inputs a certain sample amount of aircraft generator bearing vibration signals to construct generator bearing health index data; the specific method is as follows Figure 2 As shown:

[0034] The vibration data eigenvalue calculation window is set to 12800 data lengths, the data window is moved every 6400 data lengths and the eigenvalue is calculated. The effective value RMS is used to calculate the eigenvalue as the health indicator data of the generator bearing.

[0035]

[0036] S1 calculates the probability density corresponding to the health indicator data based on the kernel density estimator (KDE), selects the health indicator data point with the largest probability density (MaximumPoint, MP), and then constructs the control line (ControlLine, CL) of the health indicator time series data based on the MP point;

[0037] S2 uses the LER (LastExitRule) module to find the LEP (LastExitPoint) data point in the current partition window, and determines whether the LEP meets the conditions of the division point (DP) according to the judgment rule. If the DP point is found, the LEP point is used as the starting point for partitioning. The specific method is as follows:

[0038] Set the partition window L and data increment ΔL of the health indicator data, and the initial partition window L is set to ΔL;

[0039] If a data point in the window and the M data points behind it (M is the set value) are all greater than the control line, then the data point is taken as the LEP data point;

[0040] Expand the partition window backward by ΔL and find the LEP data points in the new window;

[0041] Determine whether the two LEP data points are the same. If they are different, continue to expand ΔL backward until a state demarcation point that meets the rules is found. If they are the same, the first LEP data point is considered to be the demarcation point, and then the partition window L is reset with the demarcation point as the starting point and the next demarcation point is found in the same way.

[0042] S3 repeats steps S0 to S2 until all data points are traversed.

[0043] like Figure 3 As shown in the figure, by analyzing the characteristics of the degradation data of the whole life cycle, the degradation data at different stages show the characteristics of stability and trend, and the local degradation process has the characteristics of step transition: the degradation data with stability presents the form of normal distribution, and the degradation data with trend presents the form of uniform distribution. Therefore, by setting the kernel function to Gaussian function, setting the adaptive bandwidth to Scott method, setting the number of discrete points to 200, constructing a KDE estimator and estimating the kernel density of the health indicator data, selecting the data point with the largest kernel density, and constructing a horizontal line through this point on the time series data graph of the health indicator. The specific method is as follows:

[0044] By setting the kernel function, adaptive bandwidth method, and discrete points, a KDE estimator is constructed and the kernel density of health indicator data is estimated;

[0045] Sort the health indicator data points from large to small according to the kernel density, and select the data point with the largest kernel density;

[0046] A horizontal line is constructed through this point on the time series data graph of the health indicator as the CL control line.

[0047] like Figure 4 As shown in the figure, the trend and stability of the data in each stage segmented by KDE are significantly different. In order to verify that the KDE segmentation method can accurately divide the degradation data in each stage, the trend degradation data is tested based on Kendall correlation analysis, and the stability degradation data is tested based on ADF and p-value. As shown in the following table, the trend and stability of the degradation data are quantitatively and qualitatively evaluated.

[0048] Table 1 Trend and stationarity test of degradation data

[0049] stage Data Index Trend-Qualitative <![CDATA[Trend - Quantitative 1 > Stationarity-Qualitative <![CDATA[Stationarity - Quantitative 2 > Seg_1 [0:546] Not significant 0.11 Significant <0.01*** Seg_2 [547:672] Not significant 0.45 Significant <0.01*** Seg_3 [673:1054] Significant 0.96 Not significant 0.97 Seg_4 [1055:1231] Not significant 0.30 Significant <0.01*** Seg_5 [1232:1320] Significant 0.91 Not significant 0.99

[0050] 1Kendall rank correlation coefficient; 2ADF test, ***p-value<0.01**p-value<0.05*p-value<0.1

[0051] By calculating the Kendall rank correlation coefficient between the degradation data of each stage and time, the correlation coefficients of the degradation data of stage 3 and stage 5 are both greater than 0.9, with significant trends; the p-value of the ADF test of the degradation data of stage 1, stage 2 and stage 4 is less than 0.01, with significant stability. Therefore, it is reasonable to use KDE segmentation to segment multi-stage degradation data. For degradation data with stable characteristics, it is not used to build a prediction model; for degradation data of stage 3 and stage 5 with significant trends, it can be used to build a degradation prediction model.

Claims

1. A method for classifying the health status of an aircraft generator bearing, characterized in that: The following steps are involved: S0 obtains the vibration signal of the aircraft generator bearing and pre-processes it to obtain health index data; S1 determines the data escape point as the first escape point according to the distribution characteristics of the health indicator data in the set partition window; S2 extends the partition window backward by a set length, and obtains the data escape point in the expanded partition window as the second escape point; S3 determines whether the first escape point is a state dividing point based on the comparison relationship between the first escape point and the second escape point. If it is not a state dividing point, it returns to step S2. Otherwise, it resets the division window and returns to step S1 until all data points of the health indicator data are traversed to obtain the state division result.

2. The method according to claim 1, characterized in that: The step of determining the data escape point according to the distribution characteristics of the health indicator data in the set partition window includes: Calculate the probability density of each data point in the divided window, and extract the data point with the largest probability density as the health indicator data point; The first escape point is determined according to the distribution relationship of each data point in the divided window relative to the health indicator data point.

3. The method according to claim 2, characterized in that The probability density of each data point within the partitioned window is calculated using a kernel density estimator.

4. The method according to claim 2, characterized in that: The determining of the first escape point according to the distribution relationship between each data point in the divided window and the health indicator data point includes: Compare the current data point and the subsequent consecutive data points in the partition window with the health index data point. If they are all larger than the health index data point, the current data point is used as the first escape point.

5. The method according to claim 4, characterized in that Also includes: If there is no first escape point in the current partition window, the partition window is extended backward by a set length and the steps of comparing the current data point and several consecutive data points thereafter in the partition window with the size of the health indicator data points are repeated until the first escape point is determined or all data points of the health indicator data are traversed.

6. The method according to claim 1, characterized in that When resetting the split window, the current state dividing point is used as the starting point.

7. The method according to claim 1, characterized in that The health indicator data is obtained by the following method: A data window of a set length is used to slide along the vibration signal of the aircraft generator bearing with a set step length; The effective value of the vibration signal of the aircraft generator bearing in each sliding window is calculated to obtain the health index data.

8. The method according to claim 1, characterized in that It also includes the steps of dividing the results according to the status and taking different processing methods for the data at different stages.

9. The method according to claim 1, characterized in that: Different processing methods are adopted for data at different stages, including: Data from stages with stationary characteristics are not used to build prediction models; The data of the stages with trend characteristics are used to build a degradation prediction model.

Images