Aero-engine bearing fault diagnosis method based on adaptive parameter FMD

Abstract

The invention provides an aero-engine bearing fault diagnosis method based on adaptive parameter FMD, and relates to the technical field of bearing fault diagnosis. A signal collected by a sensor is decomposed through FMD, and a kurtosis-correlation coefficient (K-C) screening criterion is adopted to screen and reconstruct a modal component. Meanwhile, in order to solve the defect that input parameters of FMD decomposition need to be manually set, a comprehensive screening index composed of skewness, kurtosis and information entropy is used as a target function to optimize the number N of decomposition modes and the size L of a filter. And finally, performing envelope demodulation on the optimal parameter reconstruction signal, and extracting fault features. A simulation signal, a data center signal of a certain university, an intermediate bearing fault signal of a test bed and a main bearing fault signal of an aero-engine are adopted to carry out validity verification on the method. Experimental results show that the processed bearing vibration signal has obvious fault features, can accurately decompose fault modes, and has robustness to other interferences and noises.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to an aircraft engine bearing fault diagnosis method based on adaptive parameter FMD (Feature Mode Decomposition). Background Art

[0002] Rolling bearings are important rotating parts of aircraft engines. They are responsible for bearing and transmitting loads during operation. At the same time, rolling bearings are in extreme working conditions of high temperature, high pressure and high speed for a long time, and bearing failures often occur. If the bearing failure is not discovered and handled in time, the bearing failure will become more and more serious, affecting the working efficiency of the aircraft engine and even causing serious safety accidents. According to statistics, in various types of aviation machinery failure accidents, the failure rate of engine rotating parts is as high as 80%, of which bearing failures account for more than half. When a bearing that has already failed is running, a periodic impact signal will be generated. The vibration signal of the bearing operation can be collected by the sensor to determine whether the bearing has failed, but the bearing fault signal is often affected by the interference of environmental noise and the influence of the signal transmission path, and the fault information cannot be accurately and efficiently extracted. Therefore, in-depth research on the invention of bearing fault diagnosis based on vibration signals is very important.

[0003] Signal analysis is a key topic in the research and application of mechanical fault diagnosis. It is a tool to extract fault features and then identify fault modes. In this field, one of the most widely used inventions for bearing fault detection is vibration signal analysis. However, the main obstacle to analyzing bearing faults from vibration data is that the signal is easily submerged by background noise and mixed with mechanical vibration signals such as shafts and gears. Therefore, it is very important to design an effective signal enhancement method to reduce noise and highlight fault features.

[0004] In view of the impulse and nonlinear characteristics of rolling bearing signals, there are currently a variety of diagnostic inventions, including: adaptive variational mode decomposition (VMD) fault feature extraction invention; wavelet packet decomposition (WPD); empirical mode decomposition (EMD) and complete adaptive noise ensemble empirical mode decomposition (CEEMDAN) rolling bearing fault diagnosis invention. Although these four inventions are effective, they all have their own limitations. WPD may generate a large number of sub-bands, which may result in some decomposed sub-bands not containing useful information, increasing the complexity and computational complexity of subsequent processing. EMD decomposition is prone to end-point effects and modal aliasing problems. CEEMDAN needs to add multiple noise auxiliary signals and perform multiple EMD decompositions, which results in much higher computational costs than EMD. This is problematic for large data sets or real-time applications, and CEEMDAN does not solve the end-point effect problem. The performance of VMD depends largely on the selection of preset parameters, such as the number of modes, penalty factors, and fidelity. End-point effects may also occur when processing signal endpoints. Moreover, when the number of selected modes is inappropriate, VMD may produce over-decomposition or under-decomposition. Summary of the invention

[0005] The technical problem to be solved by the present invention is to address the deficiencies of the above-mentioned prior art and provide an aircraft engine bearing fault diagnosis method based on adaptive parameter FMD. By decomposing different modes through the designed finite impulse response filter, the fault mode can be accurately decomposed and it is robust to other interference and noise.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] An aircraft engine bearing fault diagnosis method based on adaptive parameter FMD includes the following steps:

[0008] Step 1: Input the collected original vibration signal and measure the relevant parameters of the bearing at the same time: the relevant parameters of the bearing include: contact angle α (°), bearing rotation frequency F (Hz), ball diameter d (mm), pitch diameter D (mm);

[0009] Step 2: Set the initial input parameters of FMD and the parameter optimization range;

[0010] Step 3: Use FMD to decompose the collected vibration signal of the rolling bearing and output different modal components;

[0011] Step 4: Combine the kurtosis and correlation coefficient to calculate the threshold TH of the original vibration signal and the comparison parameter B of each modal component and compare them to screen and reconstruct the modal components;

[0012] Step 5: Calculate the kurtosis-skewness-information entropy normalized comprehensive screening index of the reconstructed signal, and determine whether the comprehensive screening index is the largest. If so, output the reconstructed signal and the corresponding input parameters; if not, update the input parameters within the optimization range and return to step 3;

[0013] Step 6: Perform envelope demodulation on the reconstructed signal to extract the characteristic frequency of the bearing fault;

[0014] Step 7: Calculate the theoretical fault characteristic frequency of the bearing based on the relevant parameters of the bearing, compare the fault characteristic frequency extracted in step 6 with the theoretical fault characteristic frequency, and determine the fault type of the rolling bearing.

[0015] Furthermore, the step 4 specifically includes:

[0016] Step 4.1: Calculate the kurtosis value ku of each modal component, as shown in the following formula:

[0017]

[0018] Among them, x i is the i-th discrete point of the modal component signal, σ is the standard deviation, is the signal average value, N is the length of the signal;

[0019] Step 4.2: Calculate the correlation coefficient C(X,Y) between each modal component and the original signal:

[0020]

[0021] Where C(X,Y) is the correlation coefficient between signal X and signal Y, Cov(X,Y) is the covariance between signal X and signal Y, and σ is the standard deviation;

[0022] Step 4.3: Calculate the threshold TH of the original vibration signal of each modal component:

[0023]

[0024] Among them, |C IMF j| represents the absolute value of the correlation coefficient of the jth modal component, represents the average value of the correlation coefficients of all modal components, n represents the number of modal components, and 1 is the normalization of the original signal kurtosis to itself;

[0025] Step 4.4: Calculate the contrast parameter B of each modal component:

[0026]

[0027] Among them, ku j is the kurtosis of each modal component, ku0 is the kurtosis of the original signal;

[0028] Step 4.5: Compare the contrast parameter B of each modal component with the original signal threshold TH, select the modal components with B ≥ TH for reconstruction, and filter out other components.

[0029] Furthermore, the specific calculation method of the comprehensive screening index in step 5 is:

[0030] Step 5.1: Calculate the skewness S:

[0031]

[0032] Step 5.2: To eliminate the interference components in the information, first evaluate the noise content in the information and introduce a parameter called information entropy H(x):

[0033]

[0034] Among them, p(x i ) is the probability of occurrence of the event at the ith discrete point in the signal x; the smaller the information entropy, the less interference the fault information receives;

[0035] Step 5.3: Normalize the kurtosis, skewness, and information entropy, and convert each parameter into a probability distribution in the range of 0 to 1. The formula is as follows:

[0036]

[0037] Among them, θ i is a dimensionless parameter of the kurtosis, skewness or information entropy normalized output, y i is the kurtosis, skewness or information entropy parameter value of the modal component, y max It is the parameter with the largest absolute value of the kurtosis, skewness or information entropy parameter;

[0038] Step 5.4: After normalizing the kurtosis, skewness, and information entropy, combine them to form a comprehensive screening index λ i :

[0039] λ i =α i +β i -δ i

[0040] Among them, α i , βi , δ i are the normalized dimensionless parameters of kurtosis, skewness and information entropy, respectively.

[0041] The beneficial effects of adopting the above technical solution are: the aircraft engine bearing fault diagnosis method based on adaptive parameter FMD provided by the present invention, the selected adaptive characteristic mode decomposition, essentially decomposes different modes through the designed finite impulse response filter. It can adaptively input parameters and accurately decompose the fault mode, and is robust to other interference and noise; the kurtosis-correlation coefficient-energy ratio criterion selected by the present invention screens out the decomposed modal components, which can effectively remove irrelevant components, avoid the one-sidedness of selecting modal components through a single indicator, and achieve the purpose of suppressing background noise. The present invention can accurately and efficiently extract fault information in the vibration signal, while avoiding the disadvantage that the FMD input parameters need to be manually screened, ensuring the accuracy and precision of the final result. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of an aircraft engine bearing fault diagnosis method based on adaptive parameter FMD provided by an embodiment of the present invention;

[0043] Figure 2 A denoised envelope spectrum for simulating a bearing signal provided by an embodiment of the present invention;

[0044] Figure 3 An overall diagram of a denoised envelope spectrum of a bearing outer ring fault on a university laboratory bench provided by an embodiment of the present invention;

[0045] Figure 4 An overall diagram of a denoised envelope spectrum of a bearing inner ring fault on a university laboratory bench provided by an embodiment of the present invention;

[0046] Figure 5 A magnified view of the damage details of the intermediate bearing provided in an embodiment of the present invention; wherein FIG (5a) is the outer ring, and FIG (5b) is the inner ring;

[0047] Figure 6 An overall diagram of a denoised envelope spectrum of an intermediate bearing outer ring fault using an adaptive parameter FMD decomposition method provided by an embodiment of the present invention;

[0048] Figure 7 An overall diagram of a denoised envelope spectrum of an intermediate bearing outer ring fault after being processed using a WPD method provided in an embodiment of the present invention;

[0049] Figure 8 An overall diagram of a denoised envelope spectrum of an intermediate bearing inner ring fault after being processed using an adaptive parameter FMD method provided by an embodiment of the present invention;

[0050] Fig. 9 An overall diagram of the denoised envelope spectrum of the inner ring fault of the intermediate bearing provided in the embodiment of the present invention after being processed by the WPD method

[0051] Fig.10 An overall diagram of a denoised envelope spectrum of an outer ring fault of an aircraft engine main bearing provided by an embodiment of the present invention after being processed by an adaptive parameter FMD method;

[0052] Fig.11 This is an overall diagram of the denoised envelope spectrum of the outer ring fault of the aircraft engine main bearing provided by the embodiment of the present invention after being processed by the WPD method. DETAILED DESCRIPTION

[0053] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0054] This embodiment uses real experimental data for analysis, which is taken from a university bearing data center. The faulty bearing selected for analysis is a 6205-2RJEM SKF deep groove ball bearing. The inner and outer rings of the bearing are processed for single-site damage using electric spark technology, and the sampling frequency of the vibration data is 12000 Hz.

[0055] like Figure 1 As shown, the method of this embodiment is described as follows.

[0056] Step 1: Use a vibration sensor to collect the vibration signal of the rolling bearing and measure the relevant parameters of the bearing, including: contact angle α (°), rotation frequency F (Hz), ball diameter d (mm), and pitch diameter D (mm).

[0057] Step 2: Set the initial input parameters of FMD and the parameter optimization range.

[0058] Step 3: Decompose the collected vibration signal of the rolling bearing using FMD to output different modal components; in this embodiment, white noise is added to the original signal:

[0059] y(t)=x(t)+n(t)

[0060]

[0061] Where x(t) is the outer ring fault signal of the rolling element, and n(t) is Gaussian white noise. Set the number of sampling points Ns = 2000, the sampling frequency fs = 12000, the displacement constant A0 = 0.6, and the damping coefficient g n =0.4, natural frequency f n =4000.

[0062] Step 4: Combine the kurtosis value and correlation coefficient to screen the decomposed modal components and reconstruct the screened components.

[0063] The process of screening the decomposed components by combining kurtosis, correlation coefficient, original signal threshold and comparison parameter is as follows:

[0064] Step 4.1: Calculate the kurtosis value ku of each component as follows:

[0065]

[0066] Among them, x i is the i-th discrete point of the modal component signal, σ is the standard deviation, is the signal average value, and N is the signal length.

[0067] Step 4.2: Calculate the correlation coefficient C(X,Y) between the modal component and the original signal:

[0068]

[0069] Where C(X,Y) is the correlation coefficient between signal X and signal Y, Cov(X,Y) is the covariance between signal X and signal Y, and σ is the standard deviation.

[0070] Step 4.4: Calculate the threshold TH of the original vibration signal of the modal component:

[0071]

[0072] Among them, |C IMF j| represents the absolute value of the correlation coefficient of the jth modal component, represents the average value of the correlation coefficients of all modal components, n represents the number of modal components, and 1 is the normalization of the original signal kurtosis to itself.

[0073] Step 4.5: Calculate the contrast parameter B of the modal component:

[0074]

[0075] Among them, ku is the kurtosis of each modal component, and ku0 is the kurtosis of the original signal.

[0076] The correlation coefficient is a value between [-1,1], which quantitatively describes the correlation between X and Y. That is, the larger the correlation coefficient, the greater the correlation. When the correlation coefficient is 0, the correlation is the lowest.

[0077] The contrast parameter B of each modal component is compared with the original signal threshold TH, the modal components with B≥TH are screened out for reconstruction, and other components are filtered out.

[0078] Step 5: In order to facilitate distinction and increase visibility, the concepts of skewness and information entropy are introduced, and these two variables are normalized together with the kurtosis, and the comprehensive screening index is calculated according to a certain method.

[0079] Step 5.1: Calculate the skewness S of the component using the following formula:

[0080]

[0081] Step 5.2: Calculate the information entropy H(x) in the component, the formula is as follows:

[0082]

[0083] Among them, p(x i ) is the probability of the event occurring at the ith discrete point in the signal x. The smaller the information entropy, the less interference the fault information receives.

[0084] Step 5.3: Normalize the kurtosis, skewness, and information entropy. The formula is as follows:

[0085]

[0086] Among them, θ i is a dimensionless parameter of the kurtosis, skewness or information entropy normalized output, y i is the kurtosis, skewness or information entropy parameter value of the modal component, y max It is the parameter with the largest absolute value of the kurtosis, skewness or information entropy parameter.

[0087] Step 5.4: After normalizing the kurtosis, skewness, and information entropy, they are integrated to form a comprehensive screening index λ i :

[0088] λ i =α i +β i -δ i

[0089] Among them, α i , β i ,δ i are the normalized dimensionless parameters of kurtosis, skewness and information entropy, respectively.

[0090] Step 6: Perform envelope demodulation on the reconstructed simulation signal to extract the characteristic frequency of the bearing fault.

[0091] Step 7: Calculate the theoretical fault characteristic frequency of the bearing based on the relevant parameters of the bearing, compare the fault characteristic frequency extracted in step 6 with the theoretical fault characteristic frequency, and determine the fault type of the rolling bearing.

[0092] In this embodiment, the denoised envelope spectrum used to simulate the bearing signal is as follows: Figure 2 As shown, it can be seen that the envelope spectrum of the signal after noise reduction highlights the fault characteristic frequency and frequency doubling components.

[0093] The overall de-noised envelope spectrum of the bearing outer ring fault of a university laboratory bench is as follows: Figure 3 As shown in the figure, it can be seen that after the adaptive parameter FMD decomposition, the obtained signal envelope spectrum can extract very rich information. The fault frequency f0 is 107.812Hz, the double frequency 2f0=215.624Hz, the triple frequency 3f0=323.436Hz, the quadruple frequency 4f0=431.248Hz, the quintuple frequency 5f0=539.06Hz, the sextuple frequency 6f0=646.872Hz, the septuple frequency 7f0=754.684Hz, the octuple frequency 8f0=862.496Hz, and the ninefold frequency 9f0=970.308Hz. In the range of 0-1000Hz, it can be clearly seen that the peak value of the fault frequency is not the largest. The amplitude of its corresponding multiple frequency rises first, the triple frequency reaches the maximum, and then begins to decline. It starts to rise after the quadruple frequency, and the quintuple frequency begins to decline again, and then rises and then declines.

[0094] The denoised envelope spectrum of the inner ring fault of a bearing on a university laboratory bench is as follows: Figure 4 As shown, the transfer frequency F r =29.95, fault frequency f i =161.179, not only can the transfer frequency, fault frequency and their corresponding multiple frequencies be extracted, but also the rich modulation frequency centered on the multiple frequency and fault frequency can be extracted. r (where n = 1, 2,), fault frequency and its multiple mf i (where m = 1, 2, 3, 4, 5, 6), and a modulation frequency af centered on the fault frequency and the multiplier i +2F r (where a=1,2,3,4,5), bf i -2F r (where b = 2,3,). Through comprehensive analysis, it can be clearly identified that the fault is a fault in the inner ring of the rolling bearing: the amplitude of the characteristic frequency of the inner ring fault is increased from f i Start to descend at 5f i It reaches a local minimum at , and then rises again to 6f i , but the amplitude reaches the maximum at the characteristic frequency, so it can be concluded that the fault occurs in the inner ring of the bearing.

[0095] Figure 5 Figure (5a) and Figure (5b) are the actual fault images of the outer ring and inner ring of the experimental intermediate bearing, respectively. The denoised envelope spectrum of the outer ring fault of the intermediate bearing obtained by adaptive parameter FMD decomposition is shown in Figure 6 As shown, it can be seen that the rotation frequency and 10 times the frequency, the outer ring fault frequency and the 2 times, 3 times, and 4 times the frequency are effectively extracted. At the same time, the modulation frequencies such as f0-10F, 2f0-10F, and 3f0+10F are also extracted. The amplitude trend of the outer ring fault frequency first rises, reaches the maximum amplitude at the fault frequency, and then gradually decreases. In summary, it can be determined that the outer ring of the rolling bearing has a fault. In order to further verify the effectiveness of the method of this embodiment, Figure 7 This is the denoised envelope diagram obtained by WPD decomposition. In comparison, it can be found that the envelope diagram obtained by WPD decomposition is messy and the fault type cannot be clearly distinguished.

[0096] The denoised envelope spectrum of the adaptive parameter FMD decomposition of the inner ring fault of the experimental intermediate bearing is shown in the figure below: Figure 8 As shown, the rotation frequency and the characteristic frequency of the inner ring fault of the intermediate bearing and its multiple frequency F = 5Hz, f i =89Hz, 2f i =178Hz, 3f i =287Hz, 4f i =356Hz. Compared with the theoretical calculation, there is an error of 1.61%. From the figure, it can be found that the amplitude reaches the maximum at the characteristic frequency and then gradually decreases. It can be analyzed that the inner ring is faulty. In order to further highlight the effectiveness of the method of this embodiment, Fig. 9 is the signal envelope spectrum after WPD method processing. It can be seen that the WPD method can only extract the rotation frequency and the fault frequency, and its amplitude is lower than the modulation frequency. The comparison results show that the adaptive parameter FMD method can be applied to the fault diagnosis of the intermediate bearing of the aircraft engine and is better than the wavelet packet decomposition.

[0097] The denoised envelope diagram of the adaptive parameter FMD decomposition of the outer ring fault of the aircraft engine main bearing in the experiment is shown in the figure below: Fig.10 As shown in the figure, the denoised envelope diagram of WPD decomposition is as follows Fig.11 In both figures, there is an obvious frequency conversion F = 240Hz in the processed signal envelope spectrum, and the characteristic frequency of the outer ring fault and its multiple frequency f o =2373Hz, 2f o =4748Hz, 3f o =7121Hz and rich modulation frequencies. From both figures, it can be found that the maximum amplitude is reached at the characteristic frequency. At the same time, it can be seen from the figure that the modulation frequency amplitude of the WPD envelope spectrum in the high-frequency area is high, but this makes the multiplication of the fault frequency not obvious.

[0098] In summary, it can be concluded that the adaptive parameter FMD method is superior to the wavelet packet decomposition (WPD) method in bearing fault diagnosis.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. An aircraft engine bearing fault diagnosis method based on adaptive parameter FMD, characterized by: The steps include: Step 1: Input the collected original vibration signal and measure the relevant parameters of the bearing at the same time: the relevant parameters of the bearing include: contact angle α (°), bearing rotation frequency F (Hz), ball diameter d (mm), pitch diameter D (mm); Step 2: Set the initial input parameters of FMD and the parameter optimization range; Step 3: Use FMD to decompose the collected vibration signal of the rolling bearing and output different modal components; Step 4: Combine the kurtosis and correlation coefficient to calculate the threshold TH of the original vibration signal and the comparison parameter B of each modal component and compare them to screen and reconstruct the modal components; Step 5: Calculate the kurtosis-skewness-information entropy normalized comprehensive screening index of the reconstructed signal, and determine whether the comprehensive screening index is the largest. If so, output the reconstructed signal and the corresponding input parameters; if not, update the input parameters within the optimization range and return to step 3; Step 6: Perform envelope demodulation on the reconstructed signal to extract the characteristic frequency of the bearing fault; Step 7: Calculate the theoretical fault characteristic frequency of the bearing based on the relevant parameters of the bearing, compare the fault characteristic frequency extracted in step 6 with the theoretical fault characteristic frequency, and determine the fault type of the rolling bearing.

2. The method for diagnosing aero-engine bearing faults based on adaptive parameter FMD according to claim 1, characterized in that: The step 4 specifically includes: Step 4.1: Calculate the kurtosis value ku of each modal component, as shown in the following formula: Among them, x i is the i-th discrete point of the modal component signal, σ is the standard deviation, is the signal average value, N is the length of the signal; Step 4.2: Calculate the correlation coefficient C(X,Y) between each modal component and the original signal: Where C(X,Y) is the correlation coefficient between signal X and signal Y, Cov(X,Y) is the covariance between signal X and signal Y, and σ is the standard deviation; Step 4.3: Calculate the threshold TH of the original vibration signal of each modal component: in, represents the absolute value of the correlation coefficient of the jth modal component, represents the average value of the correlation coefficients of all modal components, n represents the number of modal components, and 1 is the normalization of the original signal kurtosis to itself; Step 4.4: Calculate the contrast parameter B of each modal component: Among them, ku j is the kurtosis of each modal component, ku0 is the kurtosis of the original signal; Step 4.5: Compare the contrast parameter B of each modal component with the original signal threshold TH, select the modal components with B ≥ TH for reconstruction, and filter out other components.

3. The method for diagnosing aero-engine bearing faults based on adaptive parameter FMD according to claim 2, characterized in that: The specific calculation method of the comprehensive screening index in step 5 is: Step 5.1: Calculate the skewness S: Step 5.2: To eliminate the interference components in the information, first evaluate the noise content in the information and introduce a parameter called information entropy H(x): Among them, p(x i ) is the probability of occurrence of the event at the ith discrete point in the signal x; the smaller the information entropy, the less interference the fault information receives; Step 5.3: Normalize the kurtosis, skewness, and information entropy, and convert each parameter into a probability distribution in the range of 0 to 1. The formula is as follows: Among them, θ i is a dimensionless parameter of the normalized output of kurtosis, skewness or information entropy, y i is the kurtosis, skewness or information entropy parameter value of the modal component, y max It is the parameter with the largest absolute value of the kurtosis, skewness or information entropy parameter; Step 5.4: After normalizing the kurtosis, skewness, and information entropy, combine them to form a comprehensive screening index λ i : l i =a i +b i -d i Among them, α i , β i , δ i are the normalized dimensionless parameters of kurtosis, skewness and information entropy, respectively.

Images