Composite fault diagnosis method for aero-engine rolling bearing

Abstract

The invention discloses a composite fault diagnosis method for an aero-engine rolling bearing, and the method comprises the steps: collecting a fault vibration signal of the rolling bearing, decomposing the fault vibration information through a natural time scale decomposition method, and obtaining a plurality of component signals; according to the Euclidean distance between the evaluation vector of each component signal and the evaluation vector of the fault vibration signal, selecting two component signals with the maximum Euclidean distance and the minimum Euclidean distance from the component signals; determining the optimal order of Wiener filtering according to the kurtosis, and inputting the component signal into Wiener filtering with the optimal order for filtering processing to obtain a filtered signal; and processing the filtering signal to obtain a signal frequency spectrum, comparing the signal frequency spectrum with the fault characteristic frequency of the bearing, and identifying the fault type corresponding to the fault vibration signal. According to the invention, noise interference can be effectively eliminated, and fault feature information in the original signal can be highlighted, so that accurate fault diagnosis of the rotating machinery can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis and analysis, and particularly to a composite fault diagnosis method for an aero-engine rolling bearing. Background Art

[0002] As an important part of large-scale mechanical equipment, the operating state of rotating machinery is closely related to the health condition of the equipment. If the damage of key components cannot be identified in time and accurately, it is very likely to cause equipment damage, resulting in huge economic losses and even casualties. Therefore, the accurate extraction of fault information of rotating machinery is of crucial importance. The working intensity of rotating machinery is very high, so it is easy to be damaged. If the location of the damage can be timely located, many accidents can be avoided. Therefore, it is very necessary to diagnose the fault types of key components of rotating machinery.

[0003] In the research on fault diagnosis of rotating machinery, the research based on signal decomposition algorithms is an important research direction. Common signal decomposition algorithms include wavelet transform, empirical mode decomposition, variational mode decomposition, etc. The intrinsic time-scale decomposition algorithm can accurately extract the dynamic characteristics in non-stationary signals, has a relatively high decomposition speed, and has a relatively wide application in vibration signal processing. After decomposing the vibration signal into component signals, regarding how to enhance the characteristics of the component signals, the current common method is to select the optimal component signal according to various indexes, and then process the selected optimal component signal. The parameter evaluation indexes of signals can measure and describe various characteristic information of signals, but a single index cannot comprehensively characterize the fault information contained in the signals. In addition, when a fault occurs in a key component of rotating machinery, due to interference such as noise, the fault characteristics in its vibration signal are relatively weak, which will affect the accuracy of fault diagnosis. In the filtering process, one of the key factors affecting the filtering effect of Wiener filtering is the selection of the input signal (i.e., the actual signal and the mixed signal), and the other key factor is the selection of the filter order.

[0004] In current engineering applications, the order value of Wiener filtering is mainly determined directly based on experience or by analyzing the relationship between different orders and other limiting conditions, or a fixed Wiener filtering order is adopted. In this way, the setting of the Wiener filtering order is inaccurate, resulting in an unsatisfactory noise reduction effect. Summary of the Invention

[0005] The purpose of the present invention is to provide a composite fault diagnosis method for an aero-engine rolling bearing, which adaptively determines the input signal of Wiener filtering by combining the intrinsic time-scale decomposition (ITD) and Euclidean distance, and determines the optimal order of Wiener filtering (WF) according to the kurtosis index, so as to improve the noise reduction effect of Wiener filtering and realize the accurate judgment of the bearing composite fault type at the same time.

[0006] To achieve the above object, the present invention proposes the following solutions: The present invention provides a composite fault diagnosis method for aero-engine rolling bearings, specifically including the following steps: S1. Collect the fault vibration signals of the rolling bearing, decompose the fault vibration information by using the intrinsic time-scale decomposition method, and obtain several component signals; S2. Construct evaluation vectors of the fault vibration signal and each component signal. According to the Euclidean distance between the evaluation vectors of each component signal and the evaluation vector of the fault vibration signal, select the component signal PRCX with the largest Euclidean distance and the component signal PRCY with the smallest Euclidean distance from each component signal; S3. Input the component signals PRCX and PRCY as input signals into the Wiener filter, determine the optimal order of the Wiener filter according to the kurtosis, and set the Wiener filter based on the optimal order; S4. Input the component signals PRCX and PRCY into the Wiener filter with the optimal order again for filtering processing to obtain a filtered signal; S5. Process the filtered signal to obtain a signal spectrum, compare the signal spectrum with the fault characteristic frequency of the bearing, and identify the fault type corresponding to the fault vibration signal.

[0007] In some specific implementation schemes, the specific process of step S1 is as follows: S11. Decompose the fault vibration signal as the signal to be decomposed, set the piecewise linear extraction operator L, and obtain the decomposed fault vibration signal: ; wherein, represents the baseline component signal, represents the intrinsic rotation component; S12. Judge whether the decomposed fault vibration signal meets the set decomposition conditions. If not, set the next signal to be decomposed as the baseline component signal in the decomposed fault signal , and repeat step S11 until the set decomposition conditions are met; S13. After decomposition according to steps S11 - S12, a fault vibration signal including multiple independent intrinsic rotation component signals and monotonic trend component signals can be obtained: ; wherein, represents the linear baseline operator, represents the intrinsic rotation component extraction operator, represents the An inherent rotational component signal, is the monotonic trend component signal obtained by iteration; S14. Use the inherent rotational component as the component signal.

[0008] In some specific implementation manners, the decomposition condition set in step S12 is whether a monotonic trend component signal appears.

[0009] In some specific implementation manners, the evaluation vector includes kurtosis, variance, and root mean square. The evaluation vector of the fault vibration signal is expressed as P = {G, I, M}. The Euclidean distance between the fault vibration signal and the evaluation vector of the signal component is calculated as follows: where, represents the Euclidean distance between the fault vibration signal and the evaluation vector of the u-th signal component. The evaluation vector of the u-th signal component is expressed as , and Z represents the total number of signal components.

[0010] In some specific implementation manners, the specific process of the filtering process in Wiener filtering is as follows: S41. Input the component signals PRCX and PRCY into Wiener filtering, use the component signal PRCX as the mixed signal, and use the component signal PRCY as the desired signal; S42. Calculate the autocorrelation function of the mixed signal and the cross-correlation function between the mixed signal and the desired signal respectively; S43. Based on the autocorrelation function and the cross-correlation function, use the least mean square error method to solve the impulse response corresponding to Wiener filtering when the mean square error is the smallest, and filter the mixed signal according to the impulse response to obtain the filtered signal.

[0011] In some specific implementation manners, the specific process of determining the optimal order in step S3 is as follows: S31. Set the current order of Wiener filtering; S32. Input the component signals PRCX and PRCY into Wiener filtering for filtering processing to obtain a noise reduction signal; S33. Calculate the kurtosis value corresponding to the noise reduction signal output by the current order, and save the current order value and the corresponding kurtosis value into a matrix at the same time; S34. Determine whether the current order is less than the preset threshold. If so, repeat steps S31 - S34; If not, find the maximum kurtosis value and its corresponding order value in the matrix, and use the order value corresponding to the maximum kurtosis value as the optimal order of Wiener filtering.

[0012] In some specific embodiments, when setting the current order of the Wiener filter, the order optimization range is set starting from the minimum order value of the order optimization range, and with a step value of 1, steps S31 - S33 are executed for each order value in the order optimization range.

[0013] In some specific embodiments, the order optimization range is 3000 - 5000.

[0014] The beneficial effects of the present invention are as follows: The present invention provides a method for filtering each component signal based on Wiener filter and applies it to the fault diagnosis of rotating machinery. By selecting kurtosis, variance, and root mean square in the parameter evaluation index to describe the fault characteristics of each component signal. This method can effectively enhance the fault characteristics in the signal and eliminate a large amount of noise interference in the signal, and finally achieve accurate identification of the fault types of key components of rotating machinery.

[0015] Among them, through the Euclidean distance between the constructed component signal and the evaluation vector with the collected fault vibration signal as the original signal, the similarity measurement between the component signal and the original signal is realized. From the perspective of the similarity between the component signal and the original signal, the input signal of the Wiener filter is adaptively selected; compared with the method of adaptively determining the input signal of the Wiener filter based on the correlation coefficient, the fault information in the signal selected by the present invention is more comprehensive, and the bearing fault characteristics extracted are more prominent; accurate identification of the bearing compound fault type is realized; The order of the Wiener filter is optimized and determined based on kurtosis. By calculating the kurtosis values of the filtered signals at different orders, the optimization and determination of the Wiener filter order are realized; compared with the conventional method (determining the Wiener filter order based on experience), the noise reduction effect of the present invention is more ideal and it is not sensitive to the bearing compound fault type, making the bearing fault characteristic frequency further prominent. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the compound fault diagnosis method for an aeroengine rolling bearing provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of the method for determining the optimal order provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of a fault vibration signal provided by an embodiment of the present invention; Figure 4 For Figure 3 the spectrum schematic diagram; Figure 5 For Figure 3 the schematic diagram of the component signals after signal decomposition; Figure 6Schematic diagram of the correspondence between kurtosis and order provided by the embodiment of the present invention; Figure 7 Time-domain diagram of the signal after Wiener filtering provided by the embodiment of the present invention; Figure 8 is Figure 7 Corresponding spectrum schematic diagram. Detailed implementation manners

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and in no way limits the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0018] Unless otherwise specifically stated, the relative arrangements of components and steps, numerical expressions and values set forth in these embodiments do not limit the scope of the present invention.

[0019] At the same time, it should be understood that, for the sake of convenience of description, the sizes of the various parts shown in the drawings are not drawn in actual proportional relationships.

[0020] In addition, for the sake of clarity and conciseness, descriptions of well-known structures, functions, and configurations may be omitted. Those of ordinary skill in the art will recognize that various changes and modifications can be made to the examples described herein without departing from the spirit and scope of the present disclosure.

[0021] For technologies, methods, and devices known to those of ordinary skill in the relevant art, detailed discussions may not be made, but where appropriate, the technologies, methods, and devices should be regarded as part of the authorization specification.

[0022] In all the examples shown and discussed here, any specific value should be construed as merely exemplary, rather than as a limitation. Therefore, other examples of the exemplary embodiments may have different values.

[0023] Embodiment 1 As Figure 1 shown, this embodiment provides a composite fault diagnosis method for an aero-engine rolling bearing, which specifically includes the following steps: S1. Collect the fault vibration signals of the rolling bearing, and decompose the fault vibration information by using the intrinsic time-scale decomposition method to obtain a plurality of component signals; The acquisition of the fault vibration signal can measure the rotational speed of the rolling bearing through the eddy current sensor in the aero-engine - rolling bearing tester. The vibration acceleration sensors installed in the horizontal and vertical directions of the bearing housing of the tester are used to measure the fault vibration signal of the bearing, and the fault vibration signal is used as the original signal for analysis. Intrinsic time-scale decomposition (ITD) is an adaptive signal decomposition method that can decompose the original vibration signal into multiple independent proper rotation components (PRCs) and a residual trend component according to the characteristics of the original vibration signal. The decomposition steps of the ITD algorithm in step S1 are as follows: S11. Take the fault vibration signal as the signal to be decomposed, set the piecewise linear extraction operator L, and obtain the decomposed fault vibration signal: ; Among them, represents the baseline component signal, represents the proper rotation component; S12. Judge whether the decomposed fault vibration signal meets the set decomposition condition. If not, set the next signal to be decomposed as the baseline component signal in the decomposed fault signal , and repeat step S11 until the set decomposition condition is met; S13. After decomposition according to steps S11 - S12, a fault vibration signal including multiple independent proper rotation component signals and a monotonic trend component signal can be obtained: ; Among them, represents the linear baseline operator, represents the proper rotation component extraction operator, then represents the th proper rotation component signal in the signal, is the monotonic trend component signal obtained by iteration; S14. Take the proper rotation component as the component signal.

[0024] S2. Construct the evaluation vectors of the fault vibration signal and each component signal. According to the Euclidean distance between the evaluation vector of each component signal and the evaluation vector of the fault vibration signal, select the component signal PRCX with the largest Euclidean distance and the component signal PRCY with the smallest Euclidean distance from each component signal; Since the smaller the value of the Euclidean distance, the closer the straight-line distance between the two vectors and the higher the similarity between the signals. The more similar a certain component after signal decomposition is to the original signal (i.e., the fault vibration signal, the same hereinafter), the greater the possibility that the component signal contains richer fault characteristic frequencies.

[0025] In order to obtain the input signal of Wiener filtering, after performing ITD on the signal, the kurtosis, variance, and root mean square of the original signal and each component signal are calculated, and the signal evaluation vector is constructed. By calculating the Euclidean distance between the evaluation vector of each component signal and the evaluation vector of the original signal, the input signal of Wiener filtering is adaptively determined. For the convenience of subsequent description, in this embodiment, and , ([[]] is the number of signal decomposition layers) respectively represent the evaluation vectors composed of the three evaluation indexes of the original signal and the component signals.

[0026] The evaluation vector constructed by the kurtosis, variance, and root mean square of the original signal is denoted as P = {G, I, M}; an evaluation vector constructed by a certain component signal , then the Euclidean distance between the evaluation vector of the original signal and the evaluation vector of this component signal is: ; By selecting the component signal PRCX with the largest Euclidean distance from the original signal and the component signal PRCY with the smallest Euclidean distance, the adaptive determination of the input signal of Wiener filtering is realized.

[0027] S3. Input the component signals PRCX and PRCY as input signals into Wiener filtering, and determine the optimal order of Wiener filtering according to kurtosis, and set Wiener filtering based on the optimal order; Wiener filtering (WF) is a filter for vibration signals in the sense of minimum mean square error. In the case of determining the input signal of Wiener filtering, the specific filtering steps are as follows: Step1: Select the component signal (PRCX) with the largest Euclidean distance between the evaluation vector and the evaluation vector of the original signal as the mixed signal containing noise ; Step2: Select the component signal (PRCY) with the smallest Euclidean distance between the evaluation vector and the evaluation vector of the original signal as the actual signal ; Denote the noise signal as , then the relationship between the mixed signal and the actual signal is: Step3: Make a linear estimate of for , and denote it as ; Step4: Calculate the error , variance , is the mean value of the signal, then: (5) According to the orthogonality principle, when the variance is the smallest and should be in an orthogonal relationship, that is should be zero, where m is a variable, and thus we can obtain: Step5: Rewrite the above formula through relevant functions, and the final required impulse response formula can be obtained as: k is a constant. If there are N observables (N is the order of Wiener filtering), then equation (7) can be written as: Writing equation (8) in matrix form gives: That is: is the impulse response in the filter, then the estimated value of the signal obtained after filtering can be expressed as: The current goal is to obtain when it is the smallest, the corresponding value.

[0028] It can be found that the process of Wiener filtering is mainly to obtain the expression of Wiener filtering coefficients under the minimum mean square error. And this only requires obtaining the autocorrelation function of the mixed signal and the cross-correlation function of the mixed signal and the actual signal , then the form of Wiener filtering can be obtained (in this embodiment, the component signals PRCX and PRCY after ITD are selected as input signals).

[0029] It can be seen from equation (8) that the filtering effect of Wiener filtering is closely related to the order. Therefore, the optimal order of Wiener filtering is optimized and determined according to kurtosis. Considering that when the order is too small or too large, the Wiener filtering effect is not significant, so after running the kurtosis values of the filtered signals when the order is in the range of 1000 - 5000, it is found that the kurtosis values are larger when the order is in the range of 3000 - 5000. Therefore, the optimized range of the order of Wiener filtering in this embodiment is selected as 3000 - 5000. The order of Wiener filtering is adaptively determined based on the maximum kurtosis value. Calculate the kurtosis value of the signal after each Wiener filtering and determine whether it is the maximum value. If it is the maximum, output the order value corresponding to the current maximum kurtosis value; otherwise, continue the search. The specific process of determining the optimal order is as Figure 2 shown: S31. Set the order optimization range of Wiener filtering to 3000 - 5000. Start setting from the minimum order value in the order optimization range, and set the current order A of Wiener filtering to 3000; S32. Input the component signals PRCX and PRCY into the Wiener filtering for filtering processing to obtain a noise-reduced signal; S33. Calculate the kurtosis value corresponding to the noise-reduced signal output at the current order, and at the same time save the current order value and the corresponding kurtosis value into a matrix; S34. Determine whether the current order is less than a preset threshold. If so, increment the current order A by 1 and repeat steps S31 - S34; If not, find the maximum kurtosis value and its corresponding order value in the matrix, and use the order value corresponding to the maximum kurtosis value as the optimal order of Wiener filtering.

[0030] S4. Input the component signals PRCX and PRCY into the Wiener filtering with the optimal order set for filtering processing to obtain a filtered signal; After determining the optimal order of Wiener filtering, the specific process of filtering processing in Wiener filtering is as follows: S41. Input the component signals PRCX and PRCY into the Wiener filtering, and use the component signal PRCX as the mixed signal and the component signal PRCY as the desired signal; S42. Calculate the autocorrelation function of the mixed signal and the cross-correlation function between the mixed signal and the desired signal respectively; S43. Based on the autocorrelation function and the cross-correlation function, use the least mean square error method to solve the impulse response corresponding to Wiener filtering when the mean square error is the smallest, and filter the mixed signal according to the impulse response to obtain a filtered signal.

[0031] S5. Process the filtered signal to obtain a signal spectrum, compare the signal spectrum with the fault characteristic frequencies of the bearing, and identify the fault type corresponding to the fault vibration signal.

[0032] When the bearing fails, characteristic frequencies corresponding to the bearing failure will appear in its vibration signal; and there is a relatively definite relationship between the fault type and the fault characteristic frequencies. Therefore, the fault type of the bearing can be judged by calculating the fault characteristic frequencies of the bearing. For example, assume that the inner ring fault characteristic frequency is , the outer ring fault characteristic frequency is , the rolling element fault characteristic frequency is , the cage fault characteristic frequency is ; the rotational speed is , and the rotation frequency is ; and use K to represent the number of bearing rolling elements, D to represent the diameter of the bearing, and d to represent the diameter of the ball. Then the calculation formulas for the characteristic frequencies corresponding to each fault are: Rotating frequency: Inner race fault characteristic frequency: Outer race fault characteristic frequency: Rolling element fault characteristic frequency: Cage fault characteristic frequency: It can be understood that the inventive concept of the present application is as follows: First, in order to effectively separate each frequency component in the vibration signal, based on the intrinsic time-scale decomposition algorithm, the vibration signal is decomposed to obtain the corresponding intrinsic rotation components. Second, three evaluation indexes, namely kurtosis, variance and root mean square, which are more sensitive to bearing faults, are selected to construct an evaluation vector to describe the fault information of the bearing. By calculating the Euclidean distance between the evaluation vector of each component signal and the evaluation vector of the original signal, the correlation degree between the component signal and the original signal is evaluated; Third, according to the strength of the correlation degree, the component signal used as the input signal of the Wiener filter is selected to solve the problem that it is difficult to accurately select the input signal of the Wiener filter; Finally, by calculating the kurtosis of the filtered signal at different orders and finding the order corresponding to the maximum kurtosis value, the order of the Wiener filter is adaptively determined, and the fault type of the key components of the rotating machinery is identified according to the spectrum of the filtered signal. Compared with the prior art, the method of filtering the component signal based on Wiener filtering proposed by the present invention can effectively eliminate noise interference and highlight the fault characteristic information in the original signal, thereby realizing the fault diagnosis of the rotating machinery and helping to realize the condition monitoring of the mechanical equipment.

[0033] For better understanding, this embodiment is directed to specific compound fault types: (1) Inner race + outer race + rolling element compound fault; (2) Outer race + inner race compound fault; (3) Inner race + rolling element compound fault. The characteristic frequencies of the bearings corresponding to each fault type are shown in Table 1.

[0034] Table 1 Characteristic frequencies of bearings corresponding to each fault mode Taking the typical data of the bearing with an outer race + rolling element compound fault as an example of the fault type, the above method is verified. The rotational speed corresponding to this state is 1534.5 r / min, and the rotational frequency is 25.6 Hz (1534.5 / 60 = 25.6). According to formulas (13)-(17), the characteristic frequencies of each fault of the bearing are calculated as = 113.4Hz, = 65.6 Hz, = 44.5 Hz, = 9.4 Hz. As Figure 3 shown, it is the acceleration signal corresponding to the fault vibration signal, Figure 4 and Figure 3 is the spectrum of. Among them, 293 Hz corresponds to the sum of 6 times the rolling element fault characteristic frequency and 1 times the rotation frequency ((293 - 25.6) / 6 = 44.6 Hz), and there are no other obvious characteristic frequency components.

[0035] For the fault vibration signal such as Figure 3 shown, after decomposing it using IDT, four component signals PRC1 - 4 as shown in Figure 5 are obtained. Calculate the Euclidean distances between each component signal and the evaluation vector of the fault vibration signal, as shown in Table 2: Table 2 Evaluation vectors and Euclidean distances between evaluation vectors Based on the minimum Euclidean distance as the selection criterion, the input signals for Wiener filtering selected are PRC1 and PRC3. According to the selected input signals and kurtosis index, the order of Wiener filtering is adaptively determined, and the result is as shown in Figure 6 shown. Filter the signal according to the adaptively determined input signals and the determined optimal order. It can be seen that the kurtosis value is the largest at 4.691, and the corresponding order is 4961. After setting the order as the end of Wiener filtering, the time - domain and frequency - domain diagrams of the signals after filtering the input signals PRC1 and PRC3 are as shown in Figures 7 - 8 shown. Analyzing Figure 8 it can be known that after Wiener filtering the signal according to the proposed method, the typical characteristic frequencies existing in the spectrum of the obtained signal are shown in Table 3: Table 3 Corresponding relationship between frequency components and bearing fault characteristic frequencies Analysis shows that the method in this paper realizes the accurate extraction of bearing compound fault characteristics and the accurate judgment of bearing compound fault types. This also proves that: by using the Euclidean distance between the constructed evaluation vector sensitive to fault characteristics and the original signal, the adaptive selection of Wiener filtering input signals is realized; the optimal order of Wiener filtering is optimized and determined based on the kurtosis index.

[0036] The above - mentioned are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Based on the technical essence of the present invention, within the spirit and principles of the present invention, any simple modifications, equivalent replacements, and improvements made to the above - mentioned embodiments still fall within the protection scope of the technical solution of the present invention.

Claims

1. A composite fault diagnosis method for aero-engine rolling bearings, characterized in that, Specifically, it includes the following steps: S1. Collect the fault vibration signals of the rolling bearing, decompose the fault vibration information by using the intrinsic time-scale decomposition method, and obtain several component signals; S2. Construct the evaluation vectors of the fault vibration signal and each component signal. According to the Euclidean distance between the evaluation vector of each component signal and the evaluation vector of the fault vibration signal, select the component signal PRCX with the largest Euclidean distance and the component signal PRCY with the smallest Euclidean distance from each component signal; S3. Take the component signals PRCX and PRCY as input signals and input them into the Wiener filter, determine the optimal order of the Wiener filter according to the kurtosis, and set the Wiener filter based on the optimal order; S4. Input the component signals PRCX and PRCY into the Wiener filter with the optimal order again for filtering processing to obtain the filtered signal; S5. Process the filtered signal to obtain the signal spectrum, compare the signal spectrum with the fault characteristic frequency of the bearing, and identify the fault type corresponding to the fault vibration signal.

2. The compound fault diagnosis method for a rolling bearing of an aeroengine according to claim 1, wherein The specific process of step S1 is as follows: S11. Take the fault vibration signal as the signal to be decomposed, set the piecewise linear extraction operator L, and obtain the decomposed fault vibration signal: ； Among them, represents the baseline component signal, represents the inherent rotation component; S12. Determine whether the decomposed fault vibration signal meets the set decomposition condition. If not, set the next signal to be decomposed as the decomposed fault signal and the baseline component signal in , and repeat step S11 until the set decomposition condition is met; S13. After decomposition according to steps S11 - S12, a fault vibration signal including multiple independent intrinsic rotation component signals and a monotonic trend component signal can be obtained: ； Among them, represents a linear baseline operator, represents an inherent rotation component extraction operator, then represents the th inherent rotation component signal in the signal, is a monotonic trend component signal obtained by iteration; S14. Take the intrinsic rotation component as the component signal.

3. A composite fault diagnosis method for a rolling bearing of an aeroengine according to claim 2, characterized in that, The decomposition condition set in step S12 is whether a monotonic trend component signal appears.

4. A composite fault diagnosis method for a rolling bearing of an aero-engine according to claim 1, characterized in that The evaluation vector includes kurtosis, variance, and root mean square. The evaluation vector of the fault vibration signal is expressed as P = {G, I, M}. The Euclidean distance is calculated as follows: ； Among them, represents the Euclidean distance between the fault vibration signal and the evaluation vector of the \(u\)-th signal component, and the evaluation vector of the \(u\)-th signal component is expressed as , where \(Z\) represents the total number of signal components.

5. A composite fault diagnosis method for aeroengine rolling bearings according to claim 1, characterized in that, The specific process of the filtering process in the Wiener filter is as follows: S41. Input the component signals PRCX and PRCY into the Wiener filter, take the component signal PRCX as the mixed signal, and take the component signal PRCY as the desired signal; S42. Calculate the autocorrelation function of the mixed signal and the cross-correlation function between the mixed signal and the desired signal respectively; S43. Based on the autocorrelation function and the cross-correlation function, use the least mean square error method to solve the impulse response corresponding to the Wiener filter when the mean square error is the smallest, and filter the mixed signal according to the impulse response to obtain the filtered signal.

6. The compound fault diagnosis method for a rolling bearing of an aeroengine according to claim 1, wherein The specific process of determining the optimal order in step S3 is as follows: S31. Set the current order of the Wiener filter; S32. Input the component signals PRCX and PRCY into the Wiener filter for filtering processing to obtain the noise-reduced signal; S33. Calculate the kurtosis value corresponding to the noise-reduced signal output at the current order, and save the current order value and the corresponding kurtosis value into the matrix at the same time; S34. Judge whether the current order is less than the preset threshold. If so, repeat steps S31 - S34; If not, find the largest kurtosis value and its corresponding order value in the matrix, and take the order value corresponding to the largest kurtosis value as the optimal order of the Wiener filter.

7. A composite fault diagnosis method for a rolling bearing of an aeroengine according to claim 6, characterized in that When setting the current order of the Wiener filter, the order optimization range starts from the smallest order value of the order optimization range, and with a step value of 1, steps S31 - S33 are executed for each order value in the order optimization range.

8. A composite fault diagnosis method for a rolling bearing of an aeroengine according to claim 7, characterized in that The order optimization range is 3000 - 5000.

Images