Gear fault evolution analysis method and system based on vibration signal morphological characteristics

Abstract

The application discloses a gear fault evolution analysis method and system based on vibration signal mode features. The gear fault evolution analysis method comprises the following steps: acquiring a vibration signal in a gear operation process, obtaining a reference spectrum line and a target spectrum line through data segmentation and spectrum preprocessing; establishing a modulation spectrum line dictionary to reconstruct the target spectrum line into a first target spectrum line, correcting the first target spectrum line into a second target spectrum line by using the reference spectrum line; differentiating a frequency domain fault component spectrum line of the second target spectrum line and the reference spectrum line, finding out an optimal frequency domain fault component of the gear operation from the frequency domain fault component spectrum line, and converting the optimal frequency domain fault component into a time domain fault mode feature. The main processing process of the vibration signal is placed on the frequency domain, and interference components are removed through shift-invariant frequency domain, so that the degree and type of the fault can be directly obtained through the extracted fault mode feature without setting a threshold value, and the gear fault evolution analysis method has the characteristics of accuracy, visualization and simple process.

Description

Technical Field

[0001] This invention relates to the field of gear transmission and condition monitoring technology, specifically to a gear fault evolution analysis method and system based on vibration signal morphological characteristics. Background Technology

[0002] Gears, as a key component in mechanical equipment for transmitting motion and power, are widely used in important fields such as aerospace, marine engineering, and transportation due to their high precision, fast speed response, and high transmission efficiency. However, due to factors such as material defects and abnormal operating conditions, gears that have been in service for a long time are prone to various types of failures, such as pitting on the tooth surface and cracks at the tooth root. These failures seriously affect the operational stability and reliability of the system.

[0003] Vibration signal analysis is one of the main methods for analyzing gear fault evolution. When a gear tooth fails, the vibration signal exhibits obvious periodic impacts in the time domain, which is reflected in the spectrum as modulation sidebands centered on the meshing frequency and spaced at the shaft rotation frequency. Traditional gear fault evolution analysis includes time-domain and frequency-domain analysis. Time-domain analysis often relies on changes in statistical parameters such as kurtosis and root mean square. Frequency-domain analysis constructs bandpass filters to extract narrowband components near the meshing frequency, or uses signal decomposition methods to decompose the original vibration signal into sub-signals, thereby extracting the fault frequency and analyzing its fault evolution patterns, such as empirical mode decomposition and variational mode decomposition. However, bandpass filtering and signal decomposition have limitations in eliminating irrelevant interference within the modulation sideband. More importantly, it is difficult to effectively distinguish gear fault types with similar spectral distributions, resulting in low accuracy. Furthermore, whether in the time domain or frequency domain, understanding the evolution of gear faults ultimately depends on changes in parameters, leading to dependence on these parameters and a lack of detailed explanation of the fault mechanism. The process is complex and not easily identifiable. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide an accurate, visual, and simple method and system for analyzing gear fault evolution based on the morphological characteristics of vibration signals.

[0005] Technical solution: The gear fault evolution analysis method based on vibration signal morphological characteristics described in this invention includes the following steps:

[0006] Vibration signals during gear operation are acquired, and baseline and target spectra are obtained through data segmentation and spectrum preprocessing.

[0007] A modulation spectral line dictionary is established to reconstruct the target spectral line as the first target spectral line, and the first target spectral line is corrected by the reference spectral line to become the second target spectral line.

[0008] Differentially extracting frequency-domain fault component spectral lines from said second target spectral line and a reference spectral line, finding the optimal frequency-domain fault component for gear operation therefrom, and converting it into time-domain fault morphology features.

[0009] Further, the vibration signal is acquired under working conditions of a gear system, and parameters of said working conditions include sampling frequency, effective frequency range, number of gear teeth, gear rotation speed, rotation frequency of a rotating shaft, and gear meshing frequency; wherein the rotation frequency of the rotating shaft is used to represent the rotation frequency of gear operation.

[0010] Further, the modulation spectral line dictionary characterizes fault frequencies by taking an attenuation coefficient as a variable, and the specific formula is as follows:

[0011]

[0012] In the formula, δ(·) represents a pulse function; α represents a modulation spectral line attenuation coefficient, 0<α<1; κ represents the number of spectral lines, 1<κ<K, according to the distribution characteristics of meshing frequency and rotation frequency of the gear system, the value of upper limit K is not greater than 0.25*z1, z1 represents the number of teeth of the target gear; ω represents an independent variable; R + represents the set of positive real numbers; N + represents the set of positive integers; f z represents the rotation frequency of the rotating shaft where the gear is located; D(ω) represents the constructed spectral line dictionary.

[0013] Further, in the process of generating the first target spectral line by the modulation spectral line dictionary, shift-invariant sparsity is introduced, and the specific process is as follows:

[0014] First, a reconstruction formula is established according to said modulation spectral line dictionary:

[0015] X an (ω)=D*s+ε

[0016] In the formula, ε represents an error term; D represents the modulation spectral line dictionary; s represents a sparse coefficient; X an (ω) represents a target spectral line;

[0017] Second, L1 regularization is introduced, the reconstruction formula is solved with a regularization coefficient as a variable and minimizing the error term as a target to obtain the first target spectral line, and the specific process is as follows:

[0018]

[0019] X ar (ω)=D*s

[0020] In the formula, argmin(·) represents taking the minimum value of the objective function; ||·||1 represents the L1 norm of the s vector; λ>0 represents a regularization parameter, which is used to control the trade-off between sparsity and the error term; Xar (ω) represents the first target spectral line; m and n represent the position of the element; N and L represent the number of rows and columns of the dictionary.

[0021] Furthermore, the process of obtaining the optimal frequency domain fault components is as follows:

[0022] The obtained frequency domain fault component spectral lines include multiple frequency domain fault components, and each frequency domain fault component corresponds to a spectral line interval. The spectral line interval of the frequency domain fault component spectral lines is extracted.

[0023] The frequency domain fault components corresponding to the spectral line spacing distributed near the rotation frequency are taken as the optimal frequency domain fault components.

[0024] Furthermore, the process of obtaining the optimal frequency domain fault components using an enumeration method is as follows:

[0025] Calculate the variance between the spectral line spacing and the rotation frequency sequentially;

[0026] With the goal of minimizing the variance, the optimal attenuation coefficient and regularization coefficient are found through enumeration, thereby obtaining the optimal first target spectral line, second target spectral line, frequency domain fault component spectral line, and spectral line spacing. The frequency domain fault component corresponding to the optimal spectral line spacing distributed near the rotation frequency is taken as the optimal frequency domain fault component.

[0027] Furthermore, correcting the first target spectral line using the reference spectral line specifically includes:

[0028] Extract the peak values ​​of the reference spectral line and the first target spectral line;

[0029] Construct a window function with the length of the rotation frequency as the minimum window length;

[0030] By aligning the spectral lines, the window function traverses the spectral range, using the reference spectral line as a guide, and corrects the frequency of the first target spectral line; when the frequency under the peak in each window is the same, the second target spectral line is obtained.

[0031] Furthermore, the data segmentation process is as follows:

[0032] The vibration signal is processed into time series data with a certain signal length;

[0033] The stationary time series data is divided and aggregated into a baseline region signal set, and the non-stationary time series data is divided and aggregated into a target region signal set; wherein, after each division, the length of the time series data contains at least one gear rotation cycle.

[0034] Furthermore, the spectrum preprocessing process is as follows:

[0035] The reference region signal set and the target region signal set are transformed from the time domain to the frequency domain based on the fast Fourier transform to obtain the reference region spectrum set and the target region spectrum set;

[0036] The spectral value sets corresponding to the reference region spectral set and the target region spectral set follow a normal distribution. The expected value of each set of spectral values ​​is used as the amplitude. After normalization, the estimated value of the frequency corresponding to each set of spectral values ​​is obtained, and the reference spectral line and the target spectral line are obtained.

[0037] The gear fault evolution analysis system based on vibration signal morphological characteristics described in this invention includes:

[0038] The spectral acquisition module is used to acquire vibration signals during gear operation, and obtains reference and target spectral lines through data segmentation and spectrum preprocessing.

[0039] The target spectral line optimization module is used to establish a modulation spectral line dictionary to reconstruct the target spectral line as a first target spectral line, and to correct the first target spectral line as a second target spectral line using a reference spectral line.

[0040] The fault feature extraction module is used to differentially extract the frequency domain fault component spectral lines of the second target spectral line and the reference spectral line, find the optimal frequency domain fault component of the gear operation, and convert it into time domain fault morphology features.

[0041] Beneficial Effects: This invention has the following significant effects: 1. Accuracy: This invention first converts the vibration signal to the frequency domain, and then performs a time-invariant frequency domain reconstruction process by constructing a modulation spectral dictionary. This eliminates interference components (including meshing frequency and noise) within the fault frequency band, resulting in a purer gear fault signal. The frequency domain sparse representation technique is only one part of the interference removal process during the reconstruction of the target spectral line into the first target spectral line. Furthermore, this invention adaptively performs various processing on the vibration signal, such as data segmentation, spectrum preprocessing, and correcting the first target spectral line to the second target spectral line from the reference spectral line. These methods effectively make the final output fault morphology features closer to the real vibration signal, improving the accuracy of feature extraction. 2. Visualization: This invention converts the frequency domain fault components near the gear's rotational frequency to the time domain, thereby obtaining intuitive fault morphology features, which, compared to traditional bandpass... Compared with filters and signal decomposition methods, this invention allows for intuitive observation of the temporal morphological changes of fault signals. The extraction process can quickly and accurately identify different fault types without setting thresholds, overcoming the problem of poor visualization of gear fault signals in traditional techniques. Furthermore, for gears with different fault degrees, it not only diagnoses single-type faults but also identifies and analyzes the evolution of faults of varying degrees, accurately reflecting their unique fault characteristics and providing an intuitive and effective tool for fault diagnosis and prediction. 3. Simple process: This invention sets all technical processes in the frequency domain, eliminating the need to build an element library in the time domain or pre-build one, thus simplifying the processing. Additionally, this invention identifies the optimal frequency domain fault components near the gear's rotational frequency from the frequency domain fault component spectrum as fault morphological features. The extraction process does not require setting thresholds, making it easier to process and more efficient. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the fault evolution analysis method.

[0043] Figure 2 The diagram shows the dynamic response under pitting failure.

[0044] Figure 3 This is a diagram showing the morphological characteristics of pitting corrosion failure.

[0045] Figure 4 Vibration signal diagrams for different degrees of pitting corrosion;

[0046] Figure 5 Characteristic diagrams of failure morphology at different pitting corrosion levels;

[0047] Figure 6 This is a schematic diagram of the dynamic response under crack failure.

[0048] Figure 7 A schematic diagram illustrating the morphological characteristics of crack failure.

[0049] Figure 8 Vibration signal diagrams for different crack intensities;

[0050] Figure 9 These are fault morphology characteristic diagrams for different degrees of cracking. Detailed Implementation

[0051] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0052] Please see Figure 1 As shown, the gear fault evolution analysis method based on vibration signal morphological characteristics described in this invention includes the following steps:

[0053] Vibration signals during gear operation are acquired, and baseline and target spectra are obtained through data segmentation and spectrum preprocessing.

[0054] A modulation spectral line dictionary is established to reconstruct the target spectral line as the first target spectral line, and the first target spectral line is corrected by the reference spectral line to become the second target spectral line.

[0055] The frequency domain fault component spectral lines of the second target spectral line and the reference spectral line are extracted by differential extraction. The optimal frequency domain fault component that best characterizes the gear operation fault is found and converted into time domain fault morphology features.

[0056] The gear fault evolution analysis system based on vibration signal morphological characteristics described in this invention includes:

[0057] The spectral acquisition module is used to acquire vibration signals during gear operation, and obtains reference and target spectral lines through data segmentation and spectrum preprocessing.

[0058] The target spectral line optimization module is used to establish a modulation spectral line dictionary to reconstruct the target spectral line as a first target spectral line, and to correct the first target spectral line as a second target spectral line using a reference spectral line.

[0059] The fault feature extraction module is used to differentially extract the frequency domain fault component spectral lines of the second target spectral line and the reference spectral line, find the optimal frequency domain fault component that best characterizes the gear operation fault, and convert it into time domain fault morphology features.

[0060] The vibration signal is acquired under the operating conditions of the gear system. The parameters of these operating conditions include the sampling frequency, effective frequency range, number of gear teeth, gear speed, shaft rotation frequency, and gear meshing frequency. The shaft rotation frequency represents the gear's operating frequency. Specifically, the sampling frequency f is determined according to the requirements of the signal acquisition system. s According to the Nyquist sampling theorem, the effective frequency range is [0, 0.5f]. sThe number of teeth z1 and z2 of the gears were determined, the rotational speed n of the gear system during the experiment was recorded, and the rotational frequency f of the shaft was further calculated. z =n / 60 and gear meshing frequency f m =f z *z1. Install acceleration sensors at key locations in the gearbox or gear transmission system, and use a signal acquisition system at a preset sampling frequency f. s Vibration signals are collected during the operation of the gears.

[0061] After acquiring the vibration signals, to facilitate subsequent processing and improve signal accuracy, the acquired vibration signals are segmented as follows:

[0062] First, the vibration signal is processed into time series data with a certain signal length. A long segment of vibration signal time series data x(t) is recorded, with a signal length of N. x For ease of explanation, assume that the signal sequence data x(t) contains both healthy and fault signals.

[0063] Subsequently, the stationary time series data is divided and aggregated into a baseline region signal set, and the non-stationary time series data is divided and aggregated into a target region signal set; wherein, after each division, the length of the time series data contains at least one gear rotation cycle. Specifically, the vibration signal number x(t) in sequence form is divided into a baseline region and a target region, where the baseline region represents the stationary portion of the vibration signal x(t) in a healthy state, and the target region represents the signal sequence of the vibration signal x(t) after the baseline region. The signals in the baseline region and the target region are respectively divided into a length equal to N. p The signal segment, and the signal length N p It should contain at least one gear rotation cycle, i.e., N p ≥z1 / f z Therefore, the reference region signal set x e (t) and target region signal set x a (t) can be expressed as:

[0064] x e (t)=[x e,1 (t),x e,2 (t),x e,3 (t),...,x e,i (t)] T (1)

[0065] x a (t)=[x a,1 (t),x a,2 (t),x a,3 (t),...,x a,j (t)]T (2)

[0066] In the formula, i and j represent x respectively. e (t) and x a The number of signal segments in (t), where i ≥ j. To ensure signal continuity, it is recommended that the signal segments in the reference region and the signal segments in the target region maintain a certain overlap when dividing the signal.

[0067] After data segmentation, to further improve signal accuracy, spectrum preprocessing is performed, as follows:

[0068] First, the reference region signal set and the target region signal set are transformed from the time domain to the frequency domain using the Fast Fourier Transform (FFT) to obtain the reference region spectrum set X. e (ω) and target region spectral set X a (ω), and retain its corresponding reference region phase spectrum set F e and the phase spectrum set F of the target region a .

[0069] Subsequently, the spectral value sets corresponding to the reference region spectral set and the target region spectral set follow a normal distribution. Using the expected value of each set of spectral values ​​as the amplitude, normalization yields an estimated frequency corresponding to each set of spectral values, thus obtaining the reference and target spectral lines. Specifically, under the premise of ensuring the continuity of the time-domain signal, the amplitude spectral set X can be considered as... e (ω) and X a (ω) spectral line values ​​at the same frequency follow a normal distribution, i.e., A e,k =[A e,k,1 A e,k,2 ,…,A e,k,i ] T and A a,k =[A a,k,1 A a,k,2 ,…,A a,k,j ] T , where k is the number of frequencies, and A e,k ~N(μ) e,k ,σ 2 e,k A a,k ~N(μ) a,k ,σ 2 a,k ). The expected frequency μ e,k and μ a,k As amplitude spectrum A respectively e,k and A a,k The estimated values ​​are obtained and normalized, thereby obtaining the estimated values ​​of k frequencies respectively. According to equations (3) and (4), the reference spectrum X is then obtained. en (ω) and target spectral line X an (ω ), a reference phase spectrum F is further obtained by the same method en and a target phase spectrum F an .

[0070] X en (ω)=nor[μ e,1 ,μ e,2 ,μ e,3 ,...,μ e,k T (3)

[0071] X an (ω)=nor[μ a,1 ,μ a,2 ,μ a,3 ,...,μ a,k T (4)

[0072] Based on the vibration mechanism of a faulty gear system, the present invention constructs a modulation spectral line dictionary, characterizes the fault frequency with the attenuation coefficient as a variable, and the specific formula is as follows:

[0073]

[0074] wherein δ(·) represents a pulse function; α represents a modulation spectral line attenuation coefficient, 0<α<1; κ represents the number of spectral lines, 1<κ<K, according to the distribution characteristics of the meshing frequency and rotation frequency of the gear system, the value of the upper limit K is not greater than 0.25*z1, z1 represents the number of teeth of the target gear; ω represents an independent variable; R + represents a set of positive real numbers; N + represents a set of positive integers; f z represents the rotation frequency of the rotating shaft where the gear is located; D(ω) represents the constructed spectral line dictionary.

[0075] In order to remove interference in the fault frequency band, shift invariance sparsity is introduced in the process of generating a first target spectral line from the modulation spectral line dictionary, and shift-invariant frequency domain reconstruction is performed on the target spectrum according to the modulation characteristics of gear faults. The specific process is as follows:

[0076] First, a reconstruction formula is established according to the modulation spectral line dictionary:

[0077] X an (ω)=D*s+ε (6)

[0078] wherein ε represents an error term; D represents the modulation spectral line dictionary; s represents a sparse coefficient; X an (ω) represents a target spectral line;

[0079] ​​Secondly, to ensure the sparsity of s, L1 regularization is introduced. Using the regularization coefficient as a variable, the reconstruction formula is solved with the goal of minimizing the error term, yielding the first target spectral line. By selecting the maximum λ value that maximizes the number of non-zero elements in the solved s weight vector, the Fast Soft Thresholding Iterative Algorithm (F-ISTA) can be chosen as the solution method. The specific process is as follows:

[0080]

[0081] X ar (ω)=D*s (8)

[0082] In the formula, argmin(·) represents minimizing the objective function; ||·||1 represents the L1 norm of the s vector; λ>0 represents the regularization parameter, used to control the trade-off between sparsity and the error term; X ar (ω) represents the first target spectral line; m and n represent the position of the element; N and L represent the number of rows and columns of the dictionary.

[0083] After obtaining the first target spectral line, considering that changes in operating conditions or machine shutdowns may occur during gear meshing, resulting in slight fluctuations in rotational speed and causing minor deviations in the effective spectral line, it is necessary to perform spectral line offset correction before feature extraction. This invention corrects the first target spectral line using the reference spectral line, specifically including:

[0084] First, the peak values ​​of the reference spectral line and the first target spectral line are extracted respectively. Taking the target spectral amplitude Aar as an example, its specific expression is:

[0085]

[0086] Secondly, a window function is constructed using the length of the rotation frequency as the minimum window length. In this step, the reconstructed target spectrum X... ar The effective frequency of (ω) includes the meshing frequency and the modulation frequency, therefore based on the reference spectrum X en (ω) for the reconstructed target spectrum X ar The (ω) spectral line is shifted. The rotation frequency f is calculated. z The corresponding length, and with a window length not greater than N. fz To construct the window function, the calculation formula is:

[0087]

[0088] By aligning spectral lines, the window function traverses the spectral range, using a reference spectral line as a guide, and corrects the frequency of the first target spectral line. When the frequency at the peak within each window is the same, the second target spectral line is obtained. Specifically, the entire spectral range is traversed, and the frequencies within each window (X) are compared. en (ω) and X arThe spectral amplitude of (ω), when the frequencies corresponding to the maximum values ​​of the two are different, for the target spectrum X ar (ω) frequency correction yields the second target spectral line X. as (ω).

[0089] After obtaining the second target spectral line, the optimal frequency domain fault component is obtained accordingly, as follows:

[0090] First, the frequency domain fault component spectral lines of the second target spectral line and the reference spectral line are extracted using differential extraction. Specifically, based on the reference spectrum X... en (ω) and the corrected target spectrum, the second target spectral line X as (ω), frequency domain fault component spectral lines are extracted by differential extraction. The obtained frequency domain fault component spectral lines include multiple frequency domain fault components, and each frequency domain fault component corresponds to a spectral line interval. The specific expression for extracting the frequency domain fault component spectral lines is shown in equation (11), where the fault component X Fault (ω) can be expressed as Its interfering component spectral lines (including meshing components and noise) X Inter (ω) can be expressed as

[0091] X f (ω)=X as (ω)-X en (ω) (11)

[0092] Next, the spectral line spacing of the frequency domain fault component spectral lines is extracted. This is achieved through the fault component X. Fault The effectiveness of fault component extraction is evaluated by the spectral line spacing Δf of (ω), which can be expressed as:

[0093] △f=[△f1,△f2,△f3,...,△f eff (12)

[0094] In the formula, the subscript eff indicates the number of spectral lines with non-zero amplitude.

[0095] Finally, the frequency domain fault components corresponding to the spectral line spacing near the rotation frequency are taken as the optimal frequency domain fault components. This invention obtains the optimal frequency domain fault components using an enumeration method, as follows:

[0096] The variance between the spectral line spacing and the rotation frequency is calculated sequentially, and then (Δf-f) is calculated. z The variance V is used to describe whether the spectral line spacing is distributed near the rotation frequency.

[0097]

[0098] With the goal of minimizing the variance V, an optimal modulation spectral dictionary is established by finding the optimal attenuation coefficient and regularization coefficient through enumeration. The optimal first target spectral line, second target spectral line, frequency domain fault component spectral line, and spectral line spacing are then generated using this dictionary. The frequency domain fault component corresponding to the spectral line spacing distributed near the rotation frequency is taken as the optimal frequency domain fault component. Specifically, since the parameter calculation is minimal, the attenuation coefficient α and regularization coefficient λ of the first target spectral line are generated through enumeration to minimize the variance V. Based on the optimal attenuation coefficient α... opt and the regularization coefficient λ opt Find the optimal frequency domain fault component X Fault_opt (ω), the inverse Fourier transform is used to further calculate the time-domain fault morphology features, and the final time-domain fault morphology features are obtained and output. Based on the extracted fault morphology features x for different time periods Fault (t) allows for a direct observation of the gear's fault evolution process. Its expression is as follows:

[0099]

[0100] Please see Figures 2 to 9 To verify the effectiveness of the method proposed in this invention, spur gears with varying degrees of surface pitting and root cracks were selected for experiments, and the results were analyzed using simulated signal data and actual vibration data. Figures 2 to 5 The data for pitting faults include the actual vibration signals acquired, the fault morphology characteristics obtained by the method of this invention, and the established dynamic response. Figures 6 to 9 The data for crack faults include the acquired actual vibration signals, fault morphology characteristics obtained by the method of this invention, and the established dynamic response. Gear faults all occurred on the pinion, with a shaft rotation speed of 1500 r / min and a sampling frequency of 12800 Hz. According to the Nyquist sampling theorem, the effective frequency range is [0, 6400]. The number of gear teeth z1 = 29 and z2 = 100 were determined. The rotational speed of the gear system during the experiment was recorded as n = 1500 r / min, and the shaft rotational frequency f was further calculated. z =25Hz and gear meshing frequency f m =f z *z1 = 725Hz. The time series data x(t) has a signal length of N. x =500000.

[0101] The main process for obtaining fault morphological characteristics is as follows:

[0102] Step 1: Data Segmentation. Divide the vibration signal data x(t) into a reference region and a target region, and then divide the signals in the reference region and the target region into segments of length N. p=100000 signal segment.

[0103] Step 2: Spectrum preprocessing.

[0104] Step 3: Reconstruct the target spectral line as the first target spectral line. In the modulation spectral dictionary construction, the attenuation coefficient α ranges from 0.2 to 0.8. K is set to 5. In the shift-invariant frequency domain reconstruction of the target spectrum, the Fast Soft Thresholding Iterative Algorithm (F-ISTA) is selected as the solution method.

[0105] Step 4: Spectral line shift correction: the first target spectral line is converted to the second target spectral line.

[0106] Step 5: Extract fault morphological features. Please refer to [link / reference]. Figure 3 , 5 7, 9, based on the extracted fault morphology characteristics x for different time periods Fault (t) allows for a direct observation of the gear's fault evolution process. By comparing the simulation results with actual signals, it can be found that the synchronization is strong and can accurately simulate actual vibration signals. Different fault morphologies are obtained for different fault states, thus enabling more accurate and convenient differentiation of fault types.

Claims

1. A method for analyzing gear fault evolution based on vibration signal morphological characteristics, characterized in that, The method includes the following steps: Vibration signals during gear operation are acquired, and baseline and target spectra are obtained through data segmentation and spectrum preprocessing. A modulation line dictionary is established to reconstruct the target spectral line as a first target spectral line, and the first target spectral line is corrected by a reference spectral line to become a second spectral line. The modulation line dictionary uses the attenuation coefficient as a variable to represent the fault frequency, and the specific formula is as follows: In the formula, δ(·) represents the pulse function; α represents the modulation spectral attenuation coefficient, 0 < α < 1; κ represents the number of spectral lines, 1 < κ < K. According to the distribution characteristics of the meshing frequency and rotation frequency of the gear system, the upper limit K is no greater than 0.25*z1, where z1 represents the number of teeth of the target gear. Indicates the independent variable; Represents the set of positive real numbers; Represents the set of positive integers; Indicates the rotational frequency of the shaft on which the gear is located; This represents the constructed spectral dictionary; In the process of reconstructing the target spectral line into the first target spectral line using the modulation spectral line dictionary, shift-invariant sparsity is introduced, and the specific process is as follows: First, a reconstruction formula is established based on the modulation spectral dictionary: In the formula, ɛ represents the error term; D represents the modulation line dictionary; s represents the sparsity coefficient; X an (ω) represents the target spectral line; Secondly, L1 regularization is introduced, and the regularization coefficient is used as a variable. The reconstruction formula is solved with the goal of minimizing the error term to obtain the first target spectral line. The specific process is as follows: In the formula, argmin(·) represents minimizing the objective function; ||·||1 represents the L1 norm of the s vector; λ>0 represents the regularization parameter, used to control the trade-off between sparsity and the error term; X ar (ω) represents the first target spectral line; m and n represent the position of the element; N and L represent the number of rows and columns of the dictionary; The step of correcting the first target spectral line to the second target spectral line from the reference spectral line specifically includes: Extract the peak values ​​of the reference spectral line and the first target spectral line; Construct a window function with the length of the rotation frequency as the minimum window length; By aligning the spectral lines, the window function traverses the spectral range, using the reference spectral line as a guide, and corrects the frequency of the first target spectral line; when the frequency under the peak in each window is the same, the second target spectral line is obtained. The frequency domain fault component spectral lines of the second target spectral line and the reference spectral line are extracted by differential extraction. The optimal frequency domain fault component of the gear operation is found from them and converted into time domain fault morphology features.

2. The gear fault evolution analysis method based on vibration signal morphological characteristics according to claim 1, characterized in that, The vibration signal is acquired under the operating conditions of the gear system. The parameters of the operating conditions include sampling frequency, effective frequency range, number of gear teeth, gear speed, rotational frequency of the shaft, and gear meshing frequency; wherein, the rotational frequency of the shaft represents the rotational frequency of the gear operation.

3. The gear fault evolution analysis method based on vibration signal morphological characteristics according to claim 2, characterized in that, The process of obtaining the optimal frequency domain fault components is as follows: The obtained frequency domain fault component spectral lines include multiple frequency domain fault components, and each frequency domain fault component corresponds to a spectral line interval. The spectral line interval of the frequency domain fault component spectral lines is extracted. The frequency domain fault components corresponding to the spectral line spacing distributed near the rotation frequency are taken as the optimal frequency domain fault components.

4. The gear fault evolution analysis method based on vibration signal morphological characteristics according to claim 3, characterized in that, The process of obtaining the optimal frequency domain fault components by enumeration is as follows: Calculate the variance between the spectral line spacing and the rotation frequency sequentially; With the goal of minimizing the variance, the optimal attenuation coefficient and regularization coefficient are found through enumeration, thereby obtaining the optimal first target spectral line, second target spectral line, frequency domain fault component spectral line, and spectral line spacing. The frequency domain fault component corresponding to the optimal spectral line spacing distributed near the rotation frequency is taken as the optimal frequency domain fault component.

5. The gear fault evolution analysis method based on vibration signal morphological characteristics according to claim 1, characterized in that, The data segmentation process is as follows: The vibration signal is processed into time series data with a certain signal length; The stationary time series data is divided and aggregated into a baseline region signal set, and the non-stationary time series data is divided and aggregated into a target region signal set; wherein, after each division, the length of the time series data contains at least one gear rotation cycle.

6. The gear fault evolution analysis method based on vibration signal morphological characteristics according to claim 5, characterized in that, The spectrum preprocessing process is as follows: The reference region signal set and the target region signal set are transformed from the time domain to the frequency domain based on the fast Fourier transform to obtain the reference region spectrum set and the target region spectrum set; The spectral value sets corresponding to the reference region spectral set and the target region spectral set follow a normal distribution. The expected value of each set of spectral values ​​is used as the amplitude. After normalization, the estimated value of the frequency corresponding to each set of spectral values ​​is obtained, and the reference spectral line and the target spectral line are obtained.

7. A gear fault evolution analysis system based on vibration signal morphological characteristics, characterized in that, include: The spectral acquisition module is used to acquire vibration signals during gear operation, and obtains reference and target spectral lines through data segmentation and spectrum preprocessing. The target spectral line optimization module is used to establish a modulation spectral line dictionary to reconstruct the target spectral line as a first target spectral line, and to correct the first target spectral line as a second target spectral line using a reference spectral line. The modulation spectral dictionary uses the attenuation coefficient as a variable to characterize the fault frequency, and the specific formula is as follows: In the formula, δ(·) represents the pulse function; α represents the modulation spectral attenuation coefficient, 0 < α < 1; κ represents the number of spectral lines, 1 < κ < K. According to the distribution characteristics of the meshing frequency and rotation frequency of the gear system, the upper limit K is no greater than 0.25*z1, where z1 represents the number of teeth of the target gear. Indicates the independent variable; Represents the set of positive real numbers; Represents the set of positive integers; Indicates the rotational frequency of the shaft on which the gear is located; This represents the constructed spectral dictionary; In the process of reconstructing the target spectral line into the first target spectral line using the modulation spectral line dictionary, shift-invariant sparsity is introduced, and the specific process is as follows: First, a reconstruction formula is established based on the modulation spectral dictionary: In the formula, ɛ represents the error term; D represents the modulation line dictionary; s represents the sparsity coefficient; X an (ω) represents the target spectral line; Secondly, L1 regularization is introduced, and the regularization coefficient is used as a variable. The reconstruction formula is solved with the goal of minimizing the error term to obtain the first target spectral line. The specific process is as follows: In the formula, argmin(·) represents minimizing the objective function; ||·||1 represents the L1 norm of the s vector; λ>0 represents the regularization parameter, used to control the trade-off between sparsity and the error term; X ar (ω) represents the first target spectral line; m and n represent the positions of the elements; N and L represent the number of rows and columns in the dictionary; The step of correcting the first target spectral line to the second target spectral line from the reference spectral line specifically includes: Extract the peak values ​​of the reference spectral line and the first target spectral line; Construct a window function with the length of the rotation frequency as the minimum window length; By aligning the spectral lines, the window function traverses the spectral range, using the reference spectral line as a guide, and corrects the frequency of the first target spectral line; when the frequency under the peak in each window is the same, the second target spectral line is obtained. The fault feature extraction module is used to differentially extract the frequency domain fault component spectral lines of the second target spectral line and the reference spectral line, find the optimal frequency domain fault component of the gear operation, and convert it into time domain fault morphology features.

Images