A single-feature health factor extraction method based on CEEMD-CC

Abstract

A single-feature health factor extraction method based on CEEMD-CC, in the field of fault prediction technology. The present invention aims to solve the problem that the existing degradation state tracking method based on a single feature quantity is difficult to reveal the nonlinear characteristics of mechanical rotating parts and has poor robustness under noise interference conditions. The method of the present invention comprises: obtaining the initial vibration signal of the rotating machinery; using CEEMD to decompose and process the initial rotating machinery vibration signal to obtain the IMF component and residual of the rotating machinery vibration signal; calculating the correlation coefficient between the IMF component and the vibration signal, and selecting the IMF component with a correlation coefficient greater than a threshold for signal reconstruction; calculating the energy entropy H of the reconstructed signal e K, with energy entropy H e K is used as a health factor to characterize the health status of rotating machinery; the correlation index Corr(E,T), the monotonicity index Mon(E) and the robustness index Rob(E) are used as evaluation criteria for each characteristic quantity to evaluate the degree of tracking closeness of the degradation process of the health factor. The present invention is used for extracting a single characteristic health factor.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault prediction and specifically relates to a single-feature health factor extraction method based on CEEMD-CC. Background Art

[0002] Rotating machinery is currently widely used in various mechanical equipment and complex working environments due to its versatility. Once it is damaged, it will not only affect the normal use of the equipment, but may also cause huge economic losses and threats to personal safety. In actual engineering applications, the load of rotating machinery and the friction, damping and other factors between rotating machinery are changing all the time, resulting in the presence of nonlinear factors and noise interference characteristics in the vibration signals acquired by sensors. Therefore, obtaining useful information characterizing the health status of rotating machinery in the original vibration signal under nonlinear factors and reducing the impact of noise interference have become prerequisites for tracking the performance degradation of rotating machinery. Accurate tracking of the performance degradation process of rotating machinery also has extremely high requirements for the stability and trend of the extracted health factors throughout the life cycle.

[0003] However, under the conditions of interference factors, existing technologies find it difficult to reveal the nonlinear characteristics of mechanical rotating parts, and it is difficult to construct health factors with high stability and good robustness throughout the entire life cycle. Summary of the invention

[0004] The present invention aims to solve the problem that the existing degradation state tracking method based on a single feature quantity is difficult to reveal the nonlinear characteristics of mechanical rotating parts and has poor robustness under noise interference conditions. A single feature health factor extraction method based on CEEMD-CC is now provided.

[0005] A single-feature health factor extraction method based on CEEMD-CC, comprising:

[0006] Step 1: Obtain an initial vibration signal of the rotating machinery; use CEEMD to decompose the initial vibration signal of the rotating machinery to obtain the IMF component and residual of the vibration signal of the rotating machinery;

[0007] Step 2: Calculate the correlation coefficient between the IMF component and the vibration signal, and select the IMF component with a correlation coefficient greater than a threshold for signal reconstruction;

[0008] Step 3: Calculate the energy entropy H of the reconstructed signal e K, with energy entropy H e K is a health factor that characterizes the health status of rotating machinery;

[0009] Step 4: Use the correlation index Corr(E,T), monotonicity index Mon(E) and robustness index Rob(E) as the evaluation criteria for each characteristic quantity to evaluate the degree of tracking of the degradation process of the health factor.

[0010] Beneficial Effects

[0011] The energy entropy of the reconstructed signal is calculated as the health factor, which contains rich information. The reconstructed signal based on CEEMD-CC can remove nonlinear factors and noise interference, thereby reducing the influence of interference factors on the feature mining effect and enhancing the robustness of the constructed health factor to better characterize the health status of rotating machinery. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flow chart of the single-feature health factor extraction method based on CEEMD-CC;

[0013] Figure 2 It is the CEEMD decomposition process. DETAILED DESCRIPTION

[0014] Specific implementation method 1: The Empirical Mode Decomposition (EMD) algorithm can decompose redundant signals into a certain number of Intrinsic Mode Functions (IMFs), and each obtained Intrinsic Mode Function contains some characteristic signals of the original signal on a differential time scale. However, in the EMD algorithm, when there are intermittent components or abnormal components in the signal of complex components, the extreme points of such signals will deviate from their positions. This phenomenon of extreme point offset is called modal aliasing. The Complete Ensemble Empirical Mode Decomposition (CEEMD) overcomes the disadvantage of modal aliasing of the EMD algorithm, and adds N pairs of noises with opposite signs at each stage of decomposition, and obtains the decomposition result after 2N integrated averages. Finally, when a unique residual is obtained, the true IMF is defined as the difference between the current residual and its local average. The number of CEEMD iterations is half that of the EMD algorithm.

[0015] In statistics, the correlation coefficient (CC) is usually used to describe the degree of correlation between two variables. Due to different research objects, there are many ways to define the correlation coefficient, and the more commonly used one is the Pearson correlation coefficient. The correlation coefficient is a statistical indicator used to reflect the closeness of the correlation between variables. The correlation coefficient is calculated using the product difference method, which is also based on the deviation of the two variables from their respective mean values, and reflects the degree of correlation between the two variables by multiplying the two deviations.

[0016] Reference Figure 1 Specifically describing this embodiment, this embodiment is a single-feature health factor extraction method based on CEEMD-CC, comprising:

[0017] Step 1: Acquire a vibration signal of a rotating machine; the vibration signal of the rotating machine is acquired by a vibration sensor;

[0018] CEEMD is used to decompose the initial rotating machinery vibration signal to obtain a series of IMF components and residuals of the rotating machinery vibration signal, such as Figure 2 As shown; the specific process is:

[0019] 2.1. Add Gaussian white noise to the vibration signal X collected by the sensor:

[0020] X′=X+β k-1 F k (1)

[0021] Where X′ represents the vibration signal after adding Gaussian white noise, F k represents the Gaussian white noise added when processing the kth IMF, β k-1 represents the signal-to-noise ratio;

[0022] 2.2. Get the first residual r1:

[0023] r1=H(X′) (2)

[0024] Where H(·) represents the local mean operator;

[0025] 2.3. Calculate the first IMF:

[0026] IMF1=X-r1 (3)

[0027] 2.4. Definition of IMF2:

[0028] IMF2=r1-r2=r1-H(r1+β1F2) (4)

[0029] 2.5. Calculate the kth residual r k :

[0030] r k =H(r k-1 +β k-1 F k ) (5)

[0031] 2.6. Get the kth IMF:

[0032] IMF k =r k-1 -r k (6)

[0033] 2.7. Repeat 2.5-2.6 until the final residual R is obtained (until the Kth IMF component is decomposed, the vibration signal is decomposed into K IMF components), and then proceed to 2.8:

[0034]

[0035] 2.8, the vibration signal X is expressed as formula (8);

[0036]

[0037] Step 2: Set a threshold, calculate the correlation coefficient between each IMF component and the vibration signal, select the IMF component with a correlation coefficient greater than the threshold for signal reconstruction, and remove the nonlinear factors and noise interference in the signal; the specific process is:

[0038] Calculate the correlation coefficient CC between each IMF component and the vibration signal k :

[0039]

[0040] in, represents the mean of the kth IMF component, Represents the mean value of the vibration signal; correlation coefficient CC k The IMF components greater than the threshold are regarded as qualified IMF components, forming a qualified IMF component set Z;

[0041] Sum all qualified IMF components in the set Z to obtain the reconstructed signal;

[0042] Step 3: Calculate the energy entropy value of the reconstructed signal as a health factor to characterize the health status of the rotating machinery; the specific process is:

[0043] Calculate the energy entropy of the reconstructed signal:

[0044]

[0045] E k =∑[IMF k (t)] (13)

[0046] Among them, H e K represents the energy entropy of the signal, p k is the proportion of the energy of the kth IMF in all components, E is the sum of all energies, and K is the number of decomposed IMFs;

[0047] The energy entropy value is used as a health factor to characterize the health status of rotating machinery;

[0048] Step 4. Use correlation, monotonicity and robustness indicators as evaluation criteria for various characteristic quantities to assess how closely the degradation process of health factors is tracked. In the evaluation criteria, the correlation indicator can measure the degree of linear correlation between the characteristic parameter sequence and the time series; the monotonicity indicator can reflect the degree of monotonic increase or decrease of the characteristic parameter sequence; and the robustness indicator can describe the tolerance of the characteristic parameter sequence to abnormal factors including noise interference.

[0049] Assume that the feature sequence is E = [e(1), e(2), ..., e(k), ..., e(N)] and the time series is T = [t1, t2, ..., t k ,...,t K ],,e(t k ) means that at time t k The eigenvalue at , N represents the total time length;

[0050] First, the moving average method is used to divide the characteristic parameter sequence into two parts: the stationary trend and the random residual:

[0051] e(t k )=e T (t k )+e R (t k ) (14)

[0052] Among them, e T (t k ) indicates the stable trend of the feature; e R (t k ) represents the random margin of the feature (the rotating machinery is initially stable, but may fail later and the feature gradually increases);

[0053] The correlation index between E and T is marked as Corr(E,T), the monotonicity index and robustness index of E are marked as Mon(E) and Rob(E), respectively. The calculation process of correlation, monotonicity and robustness index is as follows:

[0054]

[0055] Where N represents the total time length, and δ(·) represents a simple unit step function;

[0056] The output health factor tracks how closely the degradation process is progressing.

Claims

1. A single-feature health factor extraction method based on CEEMD-CC, characterized in that: include: Step 1: Obtain an initial vibration signal of the rotating machinery; use CEEMD to decompose the initial vibration signal of the rotating machinery to obtain the IMF component and residual of the vibration signal of the rotating machinery; The specific process of step one is: Add Gaussian white noise to the vibration signal X: X′=X+β k-1 F k (1) Where X′ represents the signal after adding Gaussian white noise; F k represents the Gaussian white noise added when processing the kth IMF, β k-1 represents the signal-to-noise ratio; Get the first residual r1: r1=H(X′) (2) Where H(·) represents the local mean operator; Calculate the first IMF: IMF1=X-r1 (3) Define IMF2: IMF2=r1-r2=r1-H(r1+β1F2) (4) Calculate the kth residual r k : r k =H(r k-1 +β k-1 F k ) (5) Get the kth IMF: IMF k =r k-1 -r k (6) Repeat (5)-(6) until the final residual R is obtained: Then the vibration signal X is expressed as formula (8): Step 2: Calculate the correlation coefficient between the IMF component and the vibration signal, and select the IMF component with a correlation coefficient greater than a threshold for signal reconstruction; Step 3: Calculate the energy entropy H of the reconstructed signal e K, energy entropy H e K is a health factor that characterizes the health status of rotating machinery; Step 4: Use the correlation index Corr(E,T), monotonicity index Mon(E) and robustness index Rob(E) as the evaluation criteria for each characteristic quantity to evaluate the degree of tracking of the degradation process of the health factor.

2. According to claim 1, a single-feature health factor extraction method based on CEEMD-CC is characterized in that: The second step calculates the correlation coefficient between the IMF component and the initial signal, and selects the IMF component with a correlation coefficient greater than a threshold to reconstruct the signal; the specific process is: Calculate the correlation coefficient CC between each IMF component and the original signal k : in, represents the mean of the kth IMF component, Represents the mean value of the vibration signal; Correlation coefficient CC k The IMF components greater than the threshold are regarded as qualified IMF components, forming a qualified IMF component set Z; and the signal is reconstructed using all qualified IMF components in the set Z.

3. According to claim 2, a single-feature health factor extraction method based on CEEMD-CC is characterized in that: The step three calculates the energy entropy of the reconstructed signal, and uses the energy entropy as a health factor to characterize the health status of the rotating machinery; the specific process is: Calculate the energy entropy of the reconstructed signal: Energy entropy is used as a health factor to characterize the health status of rotating machinery. k It represents the proportion of the energy of the kth IMF in all components; p k Calculated by the following formula: Among them, H e K represents the energy entropy of the signal; E represents the sum of all energies, and K is the number of decomposed IMFs.

4. The single-feature health factor extraction method based on CEEMD-CC according to claim 1 is characterized in that: The fourth step uses the correlation index Corr(E,T), the monotonicity index Mon(E) and the robustness index Rob(E) as the evaluation criteria for each characteristic quantity to evaluate the degree of tracking closeness of the degradation process of the health factor; the specific process is: Assume that the feature sequence is E = [e(1), e(2), ..., e(k), ..., e(N)] and the time series is T = [t1, t2, ..., t k ,...,t N ],e(t k ) means that at time t k The eigenvalue at , N represents the total time length; Separate the characteristic parameter sequence into a stationary trend and a random residual: e(t k )=e T (t k )+e R (t k ) (14) Among them, e T (t k ) indicates the stable trend of the feature; e R (t k ) represents the random residual of the feature; The correlation index between E and T is marked as Corr(E,T), the monotonicity index and robustness index of E are marked as Mon(E) and Rob(E), respectively. The calculation process of correlation, monotonicity and robustness index is as follows: Where K represents the total time length, δ(·) represents the unit step function; The output health factor tracks how closely the degradation process is progressing.

5. The single-feature health factor extraction method based on CEEMD-CC according to claim 1 is characterized in that: The vibration signal of the rotating machine is acquired by a vibration sensor.

Images