An aviation electromechanical component composite fault extraction method based on frequency domain feature analysis

Abstract

The application discloses a kind of aviation electromechanical component composite fault extraction methods based on frequency domain feature analysis, belong to aviation electromechanical system health monitoring and fault diagnosis technical field, to solve the problem that aviation electromechanical component is under complex service condition and generates composite fault and is difficult to effectively extract.This method is first to original discrete signal short-time Fourier transform and obtains time-frequency distribution matrix;Again along the frequency spectrum frequency axis secondary sampling and executes mean removal, time frame direction FFT and normalization, constructs with frequency spectrum frequency and cycle frequency as coordinate frequency-frequency domain matrix;Subsequently in frequency-frequency domain matrix is searched locally by slice, and candidate features are screened in combination with theoretical fault frequency matching and physical rationality constraint;Finally, the background amplitude pedestal is deducted by center amplitude, and the optimal characteristic frequency of various faults is output.The interference of structure resonance and wideband background noise is reduced, and the distinguishability and feature extraction stability of composite fault source are improved.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring and fault diagnosis technology of aviation electromechanical systems, and specifically relates to a method for extracting complex faults of aviation electromechanical components based on frequency domain feature analysis. Background Technology

[0002] In modern aviation equipment, the electromechanical system plays a crucial role in support, transmission, drive, and energy conversion, and its operational status directly affects the safety and reliability of the aircraft. Within this system, key devices such as motors, pumps, actuators, fans, and transmission mechanisms contain numerous core components, including rotors, shafts, gears, impellers, ball screws, couplings, and support bearings. These components operate under conditions of high speed, variable loads, strong vibration, and complex environments, making them prone to various failure modes, including fatigue spalling, wear, cracks, eccentricity, imbalance, loosening, and localized damage.

[0003] In actual service, due to the compact structure, complex coupling relationships, and coexistence of multiple excitation sources in aircraft electromechanical systems, key components such as rotors, shafts, and bearings often exhibit complex fault modes with multiple concurrent faults. These complex faults manifest in vibration signals as the superposition of multiple impact modulation components, with fault characteristic frequencies coupling with each other. They are also affected by frequency conversion components, structural resonance, and broadband background noise, resulting in fault characteristics characterized by weak amplitude, strong interference, and susceptibility to submersion, posing a significant challenge to fault diagnosis.

[0004] For the feature extraction of complex faults in aerospace electromechanical components, traditional time-frequency analysis methods such as the short-time Fourier transform are widely used in engineering practice. However, constrained by the Heisenberg uncertainty principle, these methods suffer from energy divergence and feature ambiguity, making it difficult to accurately characterize the localization of fault features in the time-frequency domain. More importantly, existing frequency band selection and feature extraction methods are mostly based on global statistical indicators. When dealing with complex fault signals, they tend to extract the single fault component with dominant energy, failing to take into account the weaker fault features in the background noise environment. This results in the weaker fault components in the complex fault being masked by the stronger fault components. In the actual needs of fault diagnosis in aerospace electromechanical systems, the ability to accurately identify and distinguish various fault components in complex faults is of great significance for achieving predictive maintenance and ensuring the safe operation of equipment. Therefore, it is necessary to establish a refined transformation and screening mechanism for time-frequency coefficients to improve the identification capability of weak complex faults under complex operating conditions.

[0005] Based on this, the present invention proposes a method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis, in order to solve the problems existing in the prior art. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis, thereby solving the problem of effectively extracting complex faults generated by aerospace electromechanical components under complex service conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis includes the following steps: Step 1: Acquire the original vibration signal and perform a short-time Fourier transform on the original vibration signal to obtain the time-frequency distribution matrix; Step 2: Using the time-frequency distribution matrix as input, perform secondary sampling along the frequency spectrum axis at a preset sampling interval to obtain frequency spectrum slices. Then, perform mean removal, fast Fourier transform along the time frame direction, and amplitude normalization on the amplitude subsequence corresponding to each frequency spectrum slice in sequence to construct a frequency-frequency domain matrix. Step 3: In the frequency-frequency domain matrix, perform candidate fault feature search to construct a preliminary candidate feature set, and then construct a refined feature set based on the preliminary candidate feature set; Step 4: Calculate the center amplitude of each candidate feature in the refined feature set, and select the candidate cycle frequency with the largest center amplitude from the candidate features of the same type of fault as the optimal fault feature frequency output for the corresponding fault type.

[0008] In a preferred embodiment, in step 1, the time-frequency distribution matrix is: ; in, This is the time-frequency distribution matrix after mapping the original time-domain signal; This is the original vibration signal; For window functions; For time frame indexing; For frequency index of the spectrum; Index of sampling points within the window; For frame shift; For window length; The number of points in the Fourier transform; The imaginary unit satisfies .

[0009] In a preferred embodiment, step 2, which involves performing secondary sampling along the frequency spectrum axis at a preset sampling interval to obtain a frequency spectrum slice, includes: For the A frequency slice of the spectrum, with its indexed frequency. and actual spectrum frequency They are respectively: ; ; in, This is the starting index for the secondary sampling; The interval is the second sampling index; Index for frequency slices in the spectrum; This represents the number of frequency slices of the spectrum retained after secondary sampling. For the first The frequency index corresponding to each frequency slice; For the first The actual frequency spectrum corresponding to each frequency spectrum slice. The sampling frequency; The number of points in the Fourier transform.

[0010] In a preferred embodiment, in step 2, the formula for calculating the mean removal process for the amplitude subsequence corresponding to each frequency slice is as follows: ; ; in, For the first Average background amplitude of each frequency slice in the spectrum; For the first A subsequence of amplitude values ​​that varies along the time frame direction for a frequency slice of the spectrum; The magnitude subsequence after removing the mean; This represents the total number of time frames obtained after the short-time Fourier transform. For time frame index.

[0011] In a preferred embodiment, in step 2, the formula for calculating the Fast Fourier Transform along the time frame direction is: ; in, For the first The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; Index for cycle frequency; Index for frequency slices in the spectrum; The imaginary unit satisfies .

[0012] In a preferred embodiment, in step 2, the frequency-frequency domain matrix... for: ； in, The frequency-frequency domain matrix is ​​the first... The cycle frequency index, the first Normalized amplitude at each frequency slice of the spectrum; The first denominator used to find the maximum value in the normalized denominator The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; This is the index of the cycle frequency used to find the maximum value in the normalized denominator; The upper limit of the positive cycle frequency index to be reserved; To prevent tiny constants with a denominator of zero.

[0013] In a preferred embodiment, in step 3, the preliminary candidate feature set is: ; in, This is a preliminary candidate feature set; For the first Candidate cycle frequency of each candidate feature; For the first Normalized amplitude of each candidate peak; For the first The actual spectral frequencies of the candidate peaks; For the first The spectral frequency slice index corresponding to each candidate peak; Cyclic frequency index The corresponding actual cycle frequency; For the first The cycle frequency index corresponding to each candidate peak; This represents the normalized amplitude of the candidate peak in the frequency-frequency domain matrix. For the first The first frequency slice retained by amplitude sorting in each frequency spectrum slice A set of candidate peak indices.

[0014] In a preferred embodiment, step 3, the process of constructing a refined feature set based on the preliminary candidate feature set, includes: The candidate cycle frequencies in the preliminary candidate feature set are matched and filtered with the theoretical fault characteristic frequencies. The matching and filtering simultaneously satisfies the relative error threshold condition and the physical reasonableness constraint that the candidate cycle frequency is lower than its spectral frequency, resulting in a refined feature set classified by fault type. .

[0015] In a preferred embodiment, in step 3, the refining feature set is classified according to the fault type. for: ; in, For the first Refined feature set corresponding to the type of fault; Index for fault categories; For the first The theoretical failure frequency of this type of fault; This is the relative error threshold.

[0016] In a preferred embodiment, in step 4, the optimal fault characteristic frequency is: ; in, For the first The optimal fault characteristic frequency is ultimately output for each type of fault. The cycle frequency corresponding to the optimal candidate feature; To make the center amplitude The largest candidate feature index; To refine the central amplitude of each candidate point in the feature set; For the first Fault-like Refined Feature Set The set of indices of all candidate features; Index for fault categories.

[0017] Compared with existing technologies, this invention provides a method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis, which has the following advantages: This invention transforms the time-frequency distribution matrix in traditional time-frequency analysis into a frequency-frequency domain matrix, converting the fuzzy energy distribution into an explicit modulation relationship. This allows the periodic impulse components hidden in different carrier frequency bands to stand out as spectral peaks on the cyclic frequency axis, providing a high-dimensional feature space for decoupling complex faults.

[0018] This invention employs slice-by-slice local search and TopK The strategy of combining preservation and retention ensures that the significance of weak fault modulation components in the local carrier frequency band is not masked by strong fault components, thus achieving effective decoupling in strong-weak fault coupling scenarios.

[0019] This invention transforms purely data-driven candidate peaks into fault features with clear physical interpretability through dual filtering of frequency matching and physical constraints, thereby improving the accuracy of feature extraction and its anti-interference capability.

[0020] This invention achieves a unified measurement across frequency bands through the center amplitude index, reduces the false boosting of candidate peak significance by structural resonance bands or broadband noise, and makes candidate peaks in different frequency bands have fair comparability.

[0021] The method of this invention can still identify and extract different fault sources under conditions of strong noise, multiple fault coupling and weak features, and has significant technical advantages and broad engineering application prospects in the diagnosis of complex faults in aerospace electromechanical components.

[0022] It solves the problem of effectively extracting complex faults generated by aircraft electromechanical components under complex service conditions. Attached Figure Description

[0023] Figure 1 This is a flowchart of the frequency-frequency domain matrix construction and candidate feature extraction process of the present invention.

[0024] Figure 2 This is a diagram showing the feature extraction effect of an embodiment of the present invention on a composite fault dataset.

[0025] Figure 3 This is a comparison chart showing the effectiveness of the fault extraction method of the present invention and existing technologies in extracting composite fault features from a composite fault dataset.

[0026] exist Figure 2 middle, Figure 2 (a) is the cyclic frequency spectrum of the outer ring fault characteristic frequency BPFO extracted in an embodiment of the present invention; Figure 2 (b) is the cyclic frequency spectrum of the inner ring fault characteristic frequency BPFI extracted in the embodiment of the present invention.

[0027] exist Figure 3 middle, Figure 3 (a) Comparison of composite fault feature frequency extraction results using the existing Fast SC method; Figure 3 (b) Comparison of composite fault feature frequency extraction results using the existing IESCFFOgram method; Figure 3 (c) is a comparison of the composite fault feature frequency extraction results of the existing SAM method. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] As per the instruction manual Figure 1 -Appendix Figure 3 As shown, this embodiment of the invention proposes a method for extracting composite faults in aerospace electromechanical components based on frequency domain feature analysis. This embodiment uses composite fault signals from a rolling bearing accelerated life test dataset as the verification object, selecting a bearing vibration data segment that simultaneously contains inner and outer ring faults for method verification. The experimental bearing model is... LDK UER204 The spindle speed is r / min, corresponding frequency Hz; sampling frequency Hz; the length of the analyzed signal is The corresponding analysis time is approximately [number] points. s. Based on the bearing geometry parameters, the number of rolling elements is... Z Equal to 8, rolling element diameter d Equal to 7.92 mm, bearing pitch diameter D Equal to 34.55 mm, contact angle β The value is 0°. Based on the above parameters, the theoretical outer ring fault characteristic frequency is approximately 107.91 Hz, and the theoretical inner ring fault characteristic frequency is approximately 172.09 Hz. Since this embodiment verifies a combined outer and inner ring fault, the above two theoretical fault frequencies are used as subsequent candidate frequency matching objects.

[0030] Based on the above operating conditions, the specific steps of the aviation electromechanical component composite fault extraction method based on frequency domain feature analysis described in this embodiment include the following when performing fault extraction: Step 1: Acquire the original vibration signal and perform a short-time Fourier transform on the original vibration signal to obtain the time-frequency distribution matrix; Step 1.1: To obtain the local energy distribution of the vibration signal in both time and frequency dimensions, the acquired one-dimensional vibration signal is denoted as... ; in, This is the index of the original discrete signal sampling points. To analyze signal length, .

[0031] In this embodiment, .

[0032] Step 1.2: Perform short-time Fourier transform on the vibration signal Time-frequency analysis is performed to map the original time-domain signal into a time-frequency distribution matrix. : ; in, This is the time-frequency distribution matrix after mapping the original time-domain signal; This is the original vibration signal; For window functions; For time frame indexing; For frequency index of the spectrum; Index of sampling points within the window; For frame shift; For window length; The number of points in the Fourier transform; The imaginary unit satisfies .

[0033] The time-frequency distribution matrix obtained after short-time Fourier transform It is a complex matrix with dimensions of . M × K ; in, M The total number of time frames. K This represents the number of frequency points in the spectrum.

[0034] Each element in the matrix Included in the first m The first time frame, the first k The complex time-frequency response at each frequency point in the spectrum has a real part representing the in-phase component and an imaginary part representing the quadrature component.

[0035] In this embodiment, it is preferable to set the window length. Frame shift Fast Fourier Transform Points .

[0036] With the above parameter configuration, the original time-domain signal is decomposed into local responses at different time frames and different spectral frequencies, forming a two-dimensional time-frequency distribution matrix. This time-frequency distribution matrix provides a data basis for subsequent identification of whether a certain spectral band is subject to periodic modulation due to fault impact.

[0037] Step 2: Using the time-frequency distribution matrix obtained in Step 1 As input, along the frequency spectrum axis according to a preset index interval Secondary sampling is performed, and the amplitude subsequence corresponding to each frequency slice of the spectrum is sequentially subjected to mean removal, fast Fourier transform along the time frame direction, and amplitude normalization to construct a spectrum frequency-based subsequence. With cycle frequency It is a two-dimensional coordinate frequency-frequency domain matrix (by converting time-frequency domain information into frequency-frequency domain information, deep decoupling of fault modulation components is achieved). Step 2.1: From the time-frequency distribution matrix X [ m , k Extract the periodic modulation information caused by the fault impact from the amplitude form of the time-frequency distribution matrix in step 1. Based on this, we first take its amplitude form: ; in, For the first The first time frame, the first Amplitude response at each frequency point in the spectrum; This corresponds to the complex time-frequency response; For time frame indexing; This is a frequency index for the spectrum.

[0038] Step 2.2: To reduce the computational load caused by point-by-point processing across the entire frequency band and to avoid redundant calculations due to high correlation between adjacent frequency points, calculations are performed along the frequency axis using fixed index intervals. Perform secondary sampling. Specifically: For the A frequency slice of the spectrum, with its indexed frequency. and actual spectrum frequency They are respectively: ; ; in, This is the starting index for the secondary sampling; The interval is the second sampling index; Index for frequency slices in the spectrum; This represents the number of frequency slices of the spectrum retained after secondary sampling. For the first The frequency index corresponding to each frequency slice; For the first The actual frequency spectrum corresponding to each frequency spectrum slice. The sampling frequency; The number of points in the Fourier transform.

[0039] In this embodiment, a preset secondary sampling interval is used. Starting index .

[0040] This sampling strategy allows for a reduction in the number of frequency points processed from [previous number] while ensuring the integrity of fault characteristics. K Reduce to R This significantly reduces computational complexity.

[0041] Step 2.3: Subsequently, for each selected frequency spectral slice, extract the amplitude subsequence that varies along the time frame direction. : ; in, This represents the total number of time frames obtained after the short-time Fourier transform. Indicates the first A subsequence of amplitude values ​​of a frequency slice of spectrum that varies with time frames; Amplitude matrix In the spectrum frequency index The value at that location.

[0042] For faults involving localized damage, periodic impacts can modulate certain carrier waves or resonant frequency bands, causing... It exhibits periodic fluctuations in the time frame direction that are related to the frequency of fault characteristics. Therefore, for Further cyclic frequency analysis can explicitly transform the fault modulation period hidden in the time-frequency matrix into a cyclic frequency peak, thereby enabling effective extraction of fault features.

[0043] Step 2.4: To mitigate the impact of differences in static amplitude levels and resonance background between different spectral frequency slices on candidate peak search, the mean of each amplitude subsequence is first calculated and then mean-reduced. The formula for mean-reducing is as follows: ; ; in, For the first Average background amplitude of each frequency slice in the spectrum; For the first A subsequence of amplitude values ​​that varies along the time frame direction for a frequency slice of the spectrum; The magnitude subsequence after removing the mean; This represents the total number of time frames obtained after the short-time Fourier transform. For time frame index.

[0044] Step 2.5: Subsequently, in order to convert the periodic fluctuations in the time frame direction into spectral peaks on the cyclic frequency axis, the mean-reduced amplitude subsequence is processed. Perform a Fast Fourier Transform along the time frame direction to obtain the cyclic frequency spectrum. : ; in, For the first The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; The magnitude subsequence after removing the mean; For time frame indexing; This represents the total number of time frames obtained after the short-time Fourier transform. Index for cycle frequency; Index for frequency slices in the spectrum; The imaginary unit satisfies .

[0045] Because of the time interval between adjacent time frames The equivalent sampling frequency of the amplitude time frame sequence is (number of sampling points). ， Therefore, the cyclic frequency index With actual cycle frequency The correspondence is as follows: ; In practical calculations, the positive frequency portion is usually taken and the DC term is removed, that is: ; in, For the first The actual cycle frequency corresponding to each cycle frequency index; The sampling frequency; For frame shift; This represents the total number of time frames. The upper limit of the cyclic frequency index reserved for the positive frequency portion; This is the floor function.

[0046] Step 2.6: To ensure comparability of cyclic frequency responses between different frequency slices, the cyclic frequency spectrum of each slice is normalized to obtain the frequency-frequency domain matrix. : ; in, The frequency-frequency domain matrix is ​​the first... The cycle frequency index, the first Normalized amplitude at each frequency slice of the spectrum; The first denominator used to find the maximum value in the normalized denominator The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; This is the index of the cycle frequency used to find the maximum value in the normalized denominator; The upper limit of the positive cycle frequency index to be reserved; To prevent tiny constants with a denominator of zero.

[0047] The resulting frequency-frequency domain matrix can be represented as follows: ; The physical coordinates corresponding to its matrix elements are: ; in, For the frequency-frequency domain matrix as a whole; For matrix elements; The number of dimensions for cycle frequency; This represents the number of frequency slices in the spectrum. The cycle frequency corresponding to the row of the matrix; These are the spectral frequencies corresponding to the columns of the matrix.

[0048] In this embodiment, a preset secondary sampling interval is used. .

[0049] Through the above-mentioned secondary sampling, mean removal, fast Fourier transform in the time frame direction, and normalization processing, the original time-frequency distribution matrix is ​​obtained. X [m , k [Converted to a frequency-frequency domain matrix] The frequency-frequency domain matrix Simultaneously, it retains the spectral band information of the fault modulation component and the modulation source cycle frequency information, ultimately enabling strong and weak faults in composite faults to be observed and filtered separately in the same feature space. For example... Figure 1 Step 2, shown below, details this transformation process. This matrix simultaneously preserves the spectral band information of the fault modulation component and the modulation source's cyclic frequency information. The spectral frequency axis characterizes the carrier or resonant band to which the fault modulation component is attached, while the cyclic frequency axis characterizes the repetition frequency of the fault impulse or modulation source. This construction of a two-dimensional feature space allows strong and weak faults in a composite fault to be observed and filtered separately within the same feature space, laying the foundation for subsequent fault feature extraction.

[0050] Step 3: Perform candidate fault feature search and refinement screening based on theoretical fault frequency; Step 3.1: Perform candidate fault feature search and construct a preliminary candidate feature set; In the frequency-frequency domain matrix In the middle, for each frequency slice of the spectrum, along the cyclic frequency axis Local peak detection is performed, and candidate cycle frequency peaks are extracted based on local neighborhood maxima conditions and amplitude ranking conditions. The candidate cycle frequencies are then recorded. Candidate amplitude , the frequency of the spectrum and corresponding slice index Construct a preliminary candidate feature set Specifically, this includes: Step 3.1.1: From the frequency-frequency domain matrix Extract candidate modulation peaks that may correspond to the repetition frequency of the fault impulse, and then analyze them in the frequency-frequency domain matrix. In the middle, for each frequency slice of the spectrum Along the cycle frequency axis Search for local high-amplitude peak values. Specifically: For candidate points First, it must satisfy the local neighborhood maximum condition: ; This condition ensures that the detected candidate points are indeed local maxima within their neighborhood, rather than spurious peaks caused by noise.

[0051] Step 3.1.2: For peak values ​​satisfying the local maximum condition, classify them according to amplitude. Sort by size from largest to smallest.

[0052] To preserve the weak fault modulation components that may exist in each carrier frequency band under complex fault scenarios, rather than only retaining the fault source with the largest global amplitude, this embodiment preferably retains the previous frequency components in each spectrum slice. K One peak, that is K =6, forming a preliminary candidate feature set: ; in, This is a preliminary candidate feature set; For the first Candidate cycle frequency of each candidate feature; For the first Normalized amplitude of each candidate peak; For the first The actual spectral frequencies of the candidate peaks; For the first The spectral frequency slice index corresponding to each candidate peak; Cyclic frequency index The corresponding actual cycle frequency; For the first The cycle frequency index corresponding to each candidate peak; This represents the normalized amplitude of the candidate peak in the frequency-frequency domain matrix. For the first The first frequency slice retained by amplitude sorting in each frequency spectrum slice A set of candidate peak indices.

[0053] Through "spectrum-by-spectrum frequency slice search" TopK The "amplitude preservation" method ensures that even if inner and outer faults coexist and one type of fault is significantly stronger, the modulation peak of the weak fault in the local carrier frequency band can still be preserved, thus preventing the weak fault candidate points from being completely submerged by the strong fault peaks during the global sorting stage.

[0054] Step 3.2: After the initial candidate feature set is constructed, a refined feature set is constructed based on the initial candidate feature set; Based on the structural parameters, motion parameters, or transmission relationships of the faulty components in aircraft electromechanical systems, a theoretical set of fault characteristic frequencies is calculated. Candidate cyclic frequencies from this preliminary candidate feature set are then matched and filtered against the theoretical fault characteristic frequencies. This matching and filtering process simultaneously satisfies a relative error threshold condition and a physical constraint that the candidate cyclic frequency is lower than its corresponding frequency in the spectrum, resulting in a refined feature set categorized by fault type. Specifically, this includes: Subsequently, in order to make the candidate features correspond to the actual failure mechanisms of aerospace electromechanical components, the theoretical failure frequency is calculated based on the component structural parameters.

[0055] Taking rolling bearings as an example, the characteristic frequency of outer ring failure Inner ring fault characteristic frequency characteristic frequency of rolling element failure and cage failure characteristic frequency They can be represented as: ; ; ; ; in, The characteristic frequency of the outer ring fault; The characteristic frequency of the inner ring fault; The characteristic frequency of rolling element failure; For cage failure characteristic frequencies; The number of rolling elements. The diameter of the rolling element, For bearing pitch diameter, Contact angle, For rotational frequency, other aerospace electromechanical components such as gears, impellers, rotors, or lead screws can also establish a corresponding set of theoretical fault frequencies based on meshing frequency, blade passage frequency, rotational frequency, or transmission ratio.

[0056] In this embodiment, the bearing model is LDK UER204 The preferred rolling element diameter is... mm, pitch diameter mm, number of rolling elements Frequency conversion Hz, contact angle The value is 0°. Based on the calibration results of the test platform, the theoretical outer ring fault frequency is taken as... Hz, theoretical inner ring fault frequency is taken Hz.

[0057] Since this embodiment verifies a combined outer-inner ring fault, it is preferable to... and As subsequent candidate frequency matching objects, and forming a theoretical fault frequency set: ; For the preliminary candidate feature set Any candidate cycle frequency Calculate its relationship with the theoretical fault frequency. Relative error: ; This embodiment preferably uses a relative error threshold. .in, Candidate cycle frequency With the Theoretical Fault Frequency The relative error between them; This is a set of fault categories. When... At that time, it was considered that the candidate cycle frequency was related to the first cycle frequency at the frequency matching level. The characteristics of the faults are consistent.

[0058] To further improve the physical interpretability of candidate features and eliminate low-frequency disturbances, frequency transfer leakage, or non-target carrier components, the candidate cyclic frequency must be lower than its spectral frequency. ; The physical meaning of this constraint is that localized damage to aircraft electromechanical components typically acts on structural resonance or higher-frequency carrier waves in the form of periodic impacts or amplitude modulation. Therefore, the first... Candidate cycle frequency of each candidate feature It should be lower than the first The actual spectral frequency of each candidate peak When a candidate cyclic frequency does not satisfy this relationship, the candidate point is more likely to be a result of low-frequency disturbance, spectral leakage, non-target harmonics, or frequency axis aliasing, and should be rejected.

[0059] After considering both frequency matching conditions and physical rationality constraints, a refined feature set categorized by fault type is obtained. : ; in, For the first Refined feature set corresponding to the type of fault; Index for fault categories; For the first The theoretical failure frequency of this type of fault; This is the relative error threshold.

[0060] Through this step, the candidate features must simultaneously meet three conditions: "local significance in the frequency domain", "matching the theoretical fault frequency", and "meeting the physical relationship that the fault modulation source is below the carrier frequency band".

[0061] Therefore, the method in this embodiment combines data-driven candidate peak search with physical fault frequency constraints determined by structural parameters, ultimately improving the accuracy, interpretability, and anti-interference capability of composite fault feature extraction.

[0062] Step 4: Refine the feature set The central amplitude is calculated from the candidate features. And select the center amplitude from the similar fault candidate features. The largest candidate cycle frequency is output as the optimal fault characteristic frequency for the corresponding fault type; specifically, it includes: Step 4.1: To avoid the structural resonance band or broadband background amplitude causing some non-target candidate peaks to be falsely amplified, the refined feature set obtained in Step 3 is modified. Calculate the center amplitude for each candidate point. Center amplitude The calculation formula is: ; in, Slice the spectrum of the candidate points by frequency. The average background amplitude is calculated using the following formula: ; Therefore, the central amplitude can also be written as: ; in, Slice the spectrum of the candidate points by frequency. Average background amplitude; The upper limit of the positive cycle frequency index to be reserved; Index for cycle frequency; This represents the normalized amplitude in the frequency-frequency domain matrix.

[0063] The center amplitude is obtained by measuring the candidate point amplitude. Background amplitude base of the same frequency slice of the spectrum By comparing the results, the false boosting of candidate peak significance by structural resonance band and broadband noise is reduced, so that candidate peaks from different frequency slices of the spectrum can be compared under a unified benchmark.

[0064] Step 4.2: For each type of fault In the corresponding refined feature set The candidate cyclic frequency with the largest center amplitude is selected as the optimal fault characteristic frequency for the final output of this type of fault. The calculation formula is as follows: ; in, For the first Fault-like Refined Feature Set The set of indices of all candidate features; Index for fault categories; To make the center amplitude The largest candidate feature index; To refine the central amplitude of each candidate point in the feature set; The cycle frequency corresponding to the optimal candidate feature; For the first The optimal fault characteristic frequency is the final output of the fault class.

[0065] By evaluating the center amplitude, candidate peaks in different frequency slices of the spectrum can be compared uniformly after removing their respective background bases, thereby avoiding the complete masking of weak fault sources by energy-dominant fault sources, and ultimately achieving the synchronous extraction of composite fault features such as outer ring faults and inner ring faults.

[0066] Experimental verification and result analysis: Based on the methods described in steps 1 to 4 above and the corresponding operating conditions of the aircraft electromechanical components in the embodiments, the experimental results are as follows: Figure 2 and Figure 3 As shown. Specifically: like Figure 2 As shown, after processing the composite fault signal using the method of this invention, outer ring fault features and inner ring fault features can be extracted. For the outer ring fault extraction results, as shown... Figure 2 As shown in (a), the most significant characteristic peak is formed near 109 Hz, and its overtone components can also be characterized synchronously. The overall peak is prominent and the spectral lines are clear, indicating that the method of this invention has a strong ability to identify the energy-dominant outer ring fault components. For the inner ring fault extraction results, as shown... Figure 2 As shown in (b), characteristic peaks and their harmonics associated with the inner ring fault can still be identified around 184Hz. The deviation of this frequency from the theoretical inner ring fault characteristic frequency is within the preset relative error threshold range. Although some green characteristic peaks in this result are affected by the red peak to their left, this effect mainly stems from the stronger energy of the outer ring fault component, rather than from random noise or background interference. This indicates that even in complex fault scenarios where the outer ring fault energy is significantly dominant, the method of this invention can still separate the inner ring fault features from the signal, demonstrating good complex fault resolution capability.

[0067] Comparative analysis: like Figure 3 The comparative analysis shown compares the method of this invention with the Fast SC algorithm, the IESCFFOgram algorithm, and the SAM algorithm under the same data segment and operating conditions. The comparison results show that existing comparison methods typically only clearly characterize one type of fault in complex fault scenarios, or exhibit strong interference peaks in the vicinity of the target fault frequency, making it difficult to accurately identify the other type of fault features. In contrast, the method of this invention can extract effective features of both outer and inner faults and achieve higher feature significance.

[0068] Among them, the Fast SC algorithm has a certain response near the outer ring fault features, but its ability to represent the inner ring fault features is severely insufficient; the IESCFFOgram algorithm is relatively sensitive to the energy-dominant component, but its ability to preserve weak fault features is limited; the SAM algorithm is prone to generating many non-target interference components in complex backgrounds, resulting in low feature significance. In contrast, the method of this invention uses the center amplitude to achieve a unified measure of significance across frequency bands, which can more effectively distinguish between real fault features and high-energy pseudo-features, and shows obvious superiority under strong noise and compound fault coupling conditions.

[0069] The results show that the method of this invention, by constructing a frequency-frequency domain matrix, jointly expresses the carrier frequency band information and modulation source cyclic frequency information of the fault modulation component in the same feature space. Combined with a slice-by-slice local peak search strategy, it reduces the influence of strong fault components masking weak fault components, achieving synchronous separation of strong and weak fault features under composite fault conditions. By introducing theoretical fault frequency matching screening based on component structural parameters and physical rationality constraints, it combines data-driven candidate peak search with physical fault mechanisms, improving the physical interpretability and anti-interference capability of the extracted features. By subtracting the background amplitude base using the center amplitude evaluation index, it reduces the spurious boosting effect of structural resonance bands and broadband background noise on the significance of candidate peaks, achieving a unified measurement of significance across frequency bands, enabling weak fault features to be effectively identified under a unified benchmark. This method is applicable to the extraction of composite fault features of aerospace electromechanical components, and can still identify and extract different fault sources under conditions of strong noise, multiple fault coupling, and coexistence of weak features.

[0070] It should be noted that the method of this invention is not only applicable to the fault diagnosis of rolling bearings, but can also be extended to gears, rotors, impellers, drive shafts, and other key aerospace electromechanical components. For different types of aerospace electromechanical components, corresponding theoretical fault frequency calculation models can be established based on their structural characteristics. For example, for gear transmission systems, a theoretical fault frequency set can be established based on the gear meshing frequency and its harmonic frequencies; for pumps and fans, a theoretical fault frequency set can be established based on the blade passage frequency; and for rotor systems, a theoretical fault frequency set can be established based on the rotational frequency and its harmonics. This flexibility makes the method of this invention widely applicable and can meet the complex fault diagnosis needs of various key components in aerospace electromechanical systems.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis, characterized in that, Includes the following steps: Step 1: Acquire the original vibration signal and perform a short-time Fourier transform on the original vibration signal to obtain the time-frequency distribution matrix; Step 2: Using the time-frequency distribution matrix as input, perform secondary sampling along the frequency spectrum axis at a preset sampling interval to obtain frequency spectrum slices. Then, perform mean removal, fast Fourier transform along the time frame direction, and amplitude normalization on the amplitude subsequence corresponding to each frequency spectrum slice in sequence to construct a frequency-frequency domain matrix. Step 3: In the frequency-frequency domain matrix, perform candidate fault feature search to construct a preliminary candidate feature set, and then construct a refined feature set based on the preliminary candidate feature set; Step 4: Calculate the center amplitude of each candidate feature in the refined feature set, and select the candidate cycle frequency with the largest center amplitude from the candidate features of the same type of fault as the optimal fault feature frequency output for the corresponding fault type.

2. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 1, characterized in that, In step 1, the time-frequency distribution matrix is: ； in, This is the time-frequency distribution matrix after mapping the original time-domain signal; This is the original vibration signal; For window functions; For time frame indexing; For frequency index of the spectrum; Index of sampling points within the window; For frame shift; For window length; The number of points in the Fourier transform; The imaginary unit satisfies .

3. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 1, characterized in that, Step 2, the process of performing secondary sampling along the frequency spectrum axis at a preset sampling interval to obtain a frequency spectrum slice includes: For the A frequency slice of the spectrum, with its indexed frequency. and actual spectrum frequency They are respectively: ； ； in, This is the starting index for the secondary sampling; The second sampling index interval; Index for frequency slices in the spectrum; This represents the number of frequency slices of the spectrum retained after secondary sampling. For the first The frequency index corresponding to each frequency slice; For the first The actual frequency spectrum corresponding to each frequency spectrum slice. The sampling frequency; The number of points in the Fourier transform.

4. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 1, characterized in that, In step 2, the formula for calculating the mean removal process for the amplitude subsequence corresponding to each frequency slice of the spectrum is as follows: ； ； in, For the first Average background amplitude of each frequency slice; For the first A subsequence of amplitude values ​​that varies along the time frame direction for a frequency slice of the spectrum; The magnitude subsequence after removing the mean; This represents the total number of time frames obtained after the short-time Fourier transform. For time frame index.

5. The method for extracting complex faults in aero-mechanical components based on frequency domain feature analysis as described in claim 4, characterized in that, In step 2, the formula for calculating the Fast Fourier Transform along the time frame direction is: ； in, For the first The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; Index for cycle frequency; Index for frequency slices in the spectrum; The imaginary unit satisfies .

6. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 5, characterized in that, In step 2, the frequency-frequency domain matrix for: ； in, The frequency-frequency domain matrix is ​​the first... The cycle frequency index, the first Normalized amplitude at each frequency slice of the spectrum; The first denominator used to find the maximum value in the normalized denominator The cyclic frequency spectrum corresponding to each frequency slice of the spectrum; This is the index of the cycle frequency used to find the maximum value in the normalized denominator; The upper limit of the positive cycle frequency index is reserved; To prevent tiny constants with a denominator of zero.

7. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 1, characterized in that, In step 3, the preliminary candidate feature set is as follows: ； in, This is a preliminary candidate feature set; For the first Candidate cycle frequency of each candidate feature; For the first Normalized amplitude of each candidate peak; For the first The actual spectral frequencies of the candidate peaks; For the first The spectral frequency slice index corresponding to each candidate peak; Cyclic frequency index The corresponding actual cycle frequency; For the first The cycle frequency index corresponding to each candidate peak; This represents the normalized amplitude of the candidate peak in the frequency-frequency domain matrix. For the first The first frequency slice retained by amplitude sorting in each frequency spectrum slice A set of candidate peak indices.

8. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 7, characterized in that, Step 3, the process of constructing a refined feature set based on the initial candidate feature set, includes: The candidate cycle frequencies in the preliminary candidate feature set are matched and filtered with the theoretical fault characteristic frequencies. The matching and filtering simultaneously satisfies the relative error threshold condition and the physical reasonableness constraint that the candidate cycle frequency is lower than its spectral frequency, resulting in a refined feature set classified by fault type. .

9. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 8, characterized in that, In step 3, the refined feature set is classified according to the fault type. for: ； in, For the first Refined feature set corresponding to the type of fault; Index for fault categories; For the first The theoretical failure frequency of this type of fault; This is the relative error threshold.

10. The method for extracting complex faults in aerospace electromechanical components based on frequency domain feature analysis as described in claim 9, characterized in that, In step 4, the optimal fault characteristic frequency is: ； in, For the first The optimal fault characteristic frequency is ultimately output for each type of fault. The cycle frequency corresponding to the optimal candidate feature; To make the center amplitude The largest candidate feature index; To refine the central amplitude of each candidate point in the feature set; For the first Fault-like Refined Feature Set The set of indices of all candidate features; Index for fault categories.

Images